Webpdf

EMERGING TRENDS IN ENGINEERING, SCIENCE AND TECHNOLOGY
FOR SOCIETY, ENERGY AND ENVIRONMENT
ICETEST_FM.indd 1 7/9/2018 4:25:53 PM

PROCEEDINGS OF THE INTERNATIONAL CONFERENCE IN EMERGING TRENDS IN
ENGINEERING, SCIENCE AND TECHNOLOGY (ICETEST 2018), JANUARY 18–20, 2018,
THRISSUR, KERALA, INDIA
Emerging Trends in Engineering,

Science and Technology for
Society, Energy and Environment
Editors
Rajesh Vanchipura
Associate Professor, Department of Mechanical Engineering,
Government Engineering College Trichur, Thrissur, Kerala, India
K.S. Jiji
Associate Professor, Department of Electrical Engineering,
Government Engineering College Trichur, Thrissur, Kerala, India

CRC Press/Balkema is an imprint of the Taylor & Francis Group, an informa business
© 2018 Taylor & Francis Group, London, UK
Typeset by V Publishing Solutions Pvt Ltd., Chennai, India
All rights reserved. No part of this publication or the information contained herein may
be reproduced, stored in a retrieval system, or transmitted in any form or by any means,
electronic, mechanical, by photocopying, recording or otherwise, without written prior
permission from the publisher.
Although all care is taken to ensure integrity and the quality of this publication and the
information herein, no responsibility is assumed by the publishers nor the author for any
damage to the property or persons as a result of operation or use of this publication and/or
the information contained herein.
Published by: CRC Press/Balkema

Schipholweg 107C, 2316 XC Leiden, The Netherlands
e-mail: [email protected]
www.crcpress.com – www.taylorandfrancis.com
ISBN: 978-0-8153-5760-5 (Hardback)

ISBN: 978-1-351-12414-0 (eBook)

Emerging Trends in Engineering, Science and Technology for Society,
Energy and Environment – Vanchipura & Jiji (Eds)
© 2018 Taylor & Francis Group, London, ISBN 978-0-8153-5760-5
Table of contents
Preface xv
Plenary sessions and keynote speeches xvii
Conference secretariat and committees xix
Editors xxiii
Civil Engineering Advancements for Sustainable Infrastructure

Development and Environment (CEASIDE)
Structural mechanics and materials
Studies on ambient cured geopolymer concrete 3
A.D. Sam & R.S. Deepa
Relationship between the modulus of elasticity and characteristic compressive
strength of B. subtilis impregnated concrete 7
C.M. Meera & V. Subha
Effect of copper slag as fine aggregate in high strength concrete 13
A. Sreelakshmi & G. Jeenu
Study on effectiveness of RC beam strengthening methods using CFRP 21
P. Prabhakaran & G. Joseph
Performance of nano-SiO2 admixed concrete under saline environment 29
M. Daniyal, S. Akhtar & A. Azam
Strengthening of structural elements using Fiber-Reinforced Polymer (FRP)—a review 37
B. Vijayasree & M. Soman
Dynamic soil parameter identification using measured vibration response data
of machine foundations 43
L. Jerard & S. Arun
Pushover analysis of GFRG-OGS building systems 51
K.S.R. Gouri, D. Menon & P.A. Meher
Low-velocity impact behaviour of composite laminates—a review 59
A. Austin & R.S. Priyadarsini
Engineering applications of laminated composites and various theories used
for their response analysis 67
R.K. Binia & K.K. Smitha
Hydraulics and water resources engineering

Trend analysis of rainfall projections of Teesta river basin, Sikkim using non-parametric
tests and ensemble empirical mode decomposition 79
S. Soorya, S. Adarsh & K.L. Priya

Management of beach morphological changes on the Trivandrum coast 87
M.U. Niloofar & K.P. Indulekha
Impact of climate change on stream flow 95
P. Dhanya & R. Thomas
Fluid structure interaction in transient cavitating flow in pipes—a review 101
R.M. Cherian, N. Sajikumar & K.S. Sumam
Modified roughness equations for erodible bed open channel flow 109
L. Dhanesh & K.P. Indulekha
Economic design of a surge tank: An alternative approach 117
S. Thara, K.S. Sumam & N. Sajikumar
Degradation study on coir geotextiles treated with hydrocarbon mixed soil 123
A.V. Praseeja & N. Sajikumar
Hydrodynamic behavior of Vizhinjam port using Delft3D 129
E.K.R. Shari & N. Sajikumar
Traffic and pavement engineering

Operating speed of different classes of vehicles at summit vertical curves on
two-lane non-urban roads 135
M.J. Neena, M. Harikrishna & M.V.L.R. Anjaneyulu
Comparison of level of service measures for two-lane rural highways
in mixed traffic 143
M. Sangeetha, M.V.L.R. Anjaneyulu & M. Harikrishna
Effect of rural highway geometry on driver workload: A step towards safety 149
A. Jacob, K.J. Jinesh, J. Akkara & J.P. Therattil
WebGIS enabled route planning system for tourists—a case study 157
C.B. Daniel & V.S. Manju
Comparison of demographic features, urban form and travel characteristics 165
R.S. Remjish & S. Shaheem
Integration of a multi-modal transit system for urban areas: A case study of Cochin city 171
S. Sreelekshmi & S. Shaheem
Quantitative evaluation of bus terminal using time-space analysis 179
V. Hridya & G. Geeva
Modeling of the transport of leachate contaminant in a landfill site:
A case study in Mangaluru 185
A. Divya, S. Shrihari & H. Ramesh
Critical review of water quality modeling and farm scale nutrient transport
models widely used 191
N. Nihal Pandey, V. Sharma & H.N. Udayashankar
A novel hybrid material for the trace removal of hexavalent chromium [Cr(VI)] from
contaminated water 197
A.R. Laiju & S. Sarkar
Solid waste management practices and decision-making in India 205
R. Rajesh & B.K. Bindhu
Powering India’s villages sustainably: A case study of Bihar 213
D. Kamath & A. Anctil
vi

Geomechanics and foundation engineering
Role of sodium silicate in strength development of cement treated clayey
soil admixed with composite promoter 223
K.R. Keerthi & J. Bindu
Stabilisation of Kuttanad clay by environment friendly methods—a review 229
K. Kannan, S. Gayathri, A.K. Jose, L.P. Nair, G. Das & P. Vinod
Development of critical state line concept from hypoplastic model simulations
on triaxial strength of silty sand 235
M. Akhila, K.R. Swamy & N. Sankar
Design of an embankment laterally supported with secant piles at Kuttanad 243
J. Jayamohan, N.R. Arun, K. Balan, S. Aswathy Nair, L.K. Vaishnavi,
M.S. Thampi, D.R. Renju & C. Lekshmi
Effect of anchorage of geosynthetic reinforcement on the behaviour
of reinforced foundation bed 249
A.M. Saji, A.A. Thomas, G. Sunny, J. Jayamohan & V.R. Suresh
An artificial neural network-based model for predicting the bearing capacity
of square footing on coir geotextile reinforced soil 253
D. Lal, N. Sankar & S. Chandrakaran
Interference effect of adjacent footings resting on granular beds
overlying weak soils 259
A.A. Khader, J. Jayaraj & S.R. Soorya
Influence of shape of cross section of footing on load-settlement behaviour 265
B. Anusha Nair, A. Vijayan, S. Chandni, S. Vijayan, J. Jayamohan & P. Sajith
Experimental study to determine the elastic wave velocities of marineclay 271
K. Anagha, M.N. Sandeep & K.S. Beena
Behavior of a single pile under combined and uplift loads: A review 277
D. Divya, R.B. Jiniraj & P.K. Jayasree
Reported correlations on compaction characteristics of fine grained soils
in the standard proctor test—a critical reappraisal 281
G. Sreelekshmy Pillai & P. Vinod
A study on the correlation of the shear modulus of soil with the California bearing
ratio and dynamic cone penetration value 287
S. Athira, S. Parvathy & V. Jaya
Load-settlement behaviour of footing on laterally confined soil 293
P.S. Sreedhu Potty, M. Jeenu, A. Raj, R.S. Krishna, J.B. Ralphin Rose, T.S. Amritha Varsha,
J. Jayamohan & R. Deepthi Chandran
Deformation behavior of sheet pile walls 299
H.S. Athira, V.S. Athira, F. Farhana, G.S. Gayathri, S. Reshma Babu,
N.P. Asha & P. Nair Radhika
Progressive Developments in Mechanical Engineering (PDME)

Fluid flow and fluid power systems
On the role of leading-edge tubercles in the pre-stall and post-stall characteristics
of airfoils 307
V.T. Gopinathan, R. Veeramanikandan, J.B. Ralphin Rose & V. Gokul
vii

Aerodynamic investigation of airfoil inspired HALE UAV 315
R. Saravanan, V. Vignesh, V. Venkatramasubramanian, R. Vaideeswaran & R. Naveen Rajeev
Industrial engineering and management

The usability of road traffic signboards in Kottayam 323
R. Rajesh, D.R. Gowri & N. Suhana
Development and analysis of robust neighbourhood search for flow-shop
scheduling problems with sequence dependent setup times 329
V. Jayakumar & R. Vanchipura
A real-world analysis of the impact of knowledge management on cost of quality
in construction projects 335
R. Chandran & S. Ramesh Krishnan
Maintenance strategies for realizing Industry 4.0: An overview 341
A.A. Sambrekar, C.R. Vishnu & R. Sridharan
Manufacturing technology and material science

Characterization of nanoliposomes and their modification for drug delivery 351
K.S. Athira & K.W. Ng
Investigation of compressive strength of impact damaged hybrid composite laminate 357
A. Madhan Kumar, V. Kathiresan, K. Ajithkumar, A. Hukkim Raja &
D. Balamanikandan
Effect of polarity in micro-electrical discharge machining 365
J. Boban, A. Lawrence, K.K. Manesh & L. Varghese
Characterization and Taguchi based modeling and analysis of dissimilar
TIG welded AISI 316L austenitic stainless steel-HSLA steel joints 371
P.V. Shaheer, A. Sadique & K.K. Ramachandran
Mechanical design, vibration and tribology

Behavior of stress intensity factor of semi-elliptical crack at different orientations
subjected to thermal load 381
M.B. Kumaraswamy & J.S. Basavaraja
Numerical models and computational methods

Computational Fluid Dynamics (CFD) simulation of gas-liquid-solid
three-phase fluidized bed 393
G.P. Deepak & K. Shaji
Fault diagnosis of self-aligning troughing rollers in a belt conveyor system
using an artificial neural network and Naive Bayes algorithm 401
S. Ravikumar, S. Kanagasabapathy, V. Muralidharan, R.S. Srijith & M. Bimalkumar
Other topics related to mechanical engineering

Structural analysis of midship section using finite element method 411
S. Prakash & K.K. Smitha
Experimental analysis of properties of a biolubricant derived from palm kernel oil 417
K. Sandeep & M. Sekar
Influence of perforated tabs on subsonic set control 423
D.S. Chand, D.L.V.V. Kannah, S. Thanigaiarasu & S. Elangovan
viii

Renewable energy and alternate fuels
Modification of a parabolic trough collector and its exergy analysis 435
O. Arjun, M.C. Nikhil & B. Sreejith
Analytical computation of GWP, ODP, RF number and TEWI analysis
of various R134a/R1270/R290 blends as R22 alternatives 443
S.V. Shaik & T.P. Ashok Babu
Indoor performance evaluation of a Photovoltaic Thermal (PVT) hybrid collector 451
K.K. Janishali, C. Sajith Babu & S. Gopi
Properties of biodiesel and blends: An investigative and comparative study 459
P.P. Yoosaf, C. Gopu & C.P. Sunil Kumar
Thermal sciences and transport phenomena

Computational study on two phase flow of liquid nitrogen with different
internal coatings 469
S.S. Bindu, S. Kafle, G.J. Philip & K.E. Reby Roy
An experimental investigation of cryogenic chilldown time in a polyurethane-coated
helical coil 479
J. Mohammed, P. Shyam Praveen & K.E. Reby Roy
Effect of concentrated solar radiation on reduction of pressure drop in oil pipeline 485
V.C. Midhun, K. Shaji & P.K. Jithesh
Effects of air jet on bluff body stabilized flame: Validation by simulation 493
S. Parvathi, V.P. Nithin, S. Nithin, N. Nived, P.A. Abdul Samad & C.P. Sunil Kumar
Identification of energy-intensified equipment for reliability analysis 501
C. Sanjeevy & J. Elias
International Conference on Advances in Chemical Engineering (ICAChE)

Design of Fractional Filter Fractional Order Proportional Integral Derivative
(FFFOPID) controller for higher order systems 511
R. Ranganayakulu, G. Uday Bhaskar Babu & A. Seshagiri Rao
Effect of parameters on electro-Fenton process for removal of oil and grease
from refinery wastewater 523
M. Vikramaraj, A.M. Manilal & P.A. Soloman
Conversion studies of methanol to olefin on boric acid treated Al/MCM-41 531
D. Kumar, N. Anand & A. Kedia
Degradation of diphenamid by UV/hydrogen peroxide advanced
oxidation process 537
M.S. Manju, V. Nishan Ahammed & K.P. Prasanth Kumar
Assessment of trihalomethanes in drinking water using gas chromatography 543
S.P. Aravind & P.A. Soloman
Adsorption of perchlorate using cationic modified rice husk 549
A.M. George, K.B. Radhakrishnan & A. Jayakumaran Nair
Biosorption of methyl orange from aqueous solution using cucurbita pepo leaves powder 555
M. Tukaram Bai, K. Latha, P. Venkat Rao & Y.V. Anudeep
Catalytic hydrodechlorination of 1,4-dichlorobenzene from wastewater 565
C. Megha, K. Sachithra & S.P. Kamble
ix

Controller tuning method for nonlinear conical tank system using MATLAB/Simulink 575
S. Krishnapriya & R. Anjana
Study on multi walled carbon nanotubes synthesis, modeling and applications 583
K. Krishnarchana, N. Manoj & S. Sangita Dash
Removal of nickel from aqueous solution using sargassum tenerrimum powder
(brown algae) by biosorption: Equilibrium, kinetics and thermodynamic studies 591
M. Tukaram Bai, P. Venkateswarlu & Y.V. Anudeep
Automation of IGCC power plant using Yokogawa DCS 603
K.J. Jyothir Rose, R. Anjana & L. Jenifer
Deprotonation studies on polyaniline polymethyl methacrylate blends processed
from formic acid 611
V.O. Rejini, A.H. Divya & R.S. Nair
Characterization of brushless DC motor for control valve actuation in rocket
propulsion systems 617
K.N. Ajeesh, S. Salim & R. Sujith Kumar
Sequestration of hydrogen sulfide and carbon dioxide from biogas 623
A.P. Pallan, S. Antony Raja, B. Sajeenabeevi & C.G. Varma
Development of ceramic membrane for microfiltration application in
biotechnology field 631
C. Mohit Kumar & D. Vasanth
Studies on carbon felt/phenolic composites as light weight advanced ablative
TPS for launch vehicle programmes 637
C.Y. Lincy, V. Sekkar, V. Kumar, A. Smitha Alex & S. Sahadev
Desalination studies on two single-sloped solar stills with heat-absorbing materials
as coating material on aluminum basin 643
P. Suman, D. Meghavathu, S. Veluru & Z. Hussain
Towards current induced magnetization reversal in magnetic nanowires 651
A.A. Antony, P.S. Anilkumar & V.O. Rejini
Synthesis and surfactant size regulation of nanoparticles of maghemite (γ-Fe2O3) 657
M. Ahmad Rather
Hydrothermal carbonization: A promising transformation process of biomass
into various product materials 663
M. Ahmad Rather
Electronics, Signal Processing and Communication

Engineering (E-SPACE)
Denoising of musical signals using wavelets specific for musical instruments 671
P.V. Sreelakshmi, A. Gayathri & M.S. Sinith
Low complexity encoding of M-ary QAM constellation for linear index codes 681
A. Shaju & S. Dhanasekaran
Improving myoelectric grasp recognition using empirical mode decomposition
and differential evolution based approach 687
C.K. Anusha, K. AjalBabu & N. Kunju
Automated aquaponics system 695
B. Sreelekshmi & K.N. Madhusoodanan

A review of the methods for despeckling in optical coherence tomography 701
K. Athira, K. Brijmohan, V.P. Gopi, K.K. Riyas, G. Wilson & T. Swetha
Generic object detection in image using SIFT, GIST and SURF descriptors 707
D.P. Joy, K.S. Shanthini & K.V. Priyaja
Non-destructive classification of watermelon ripeness using acoustic cues 715

R. Rajan & R.S. Reshma
Predominant instrument recognition from polyphonic music using

feature fusion 721
R. Ajayakumar & R. Rajan
A technique for countering the integer boundary spurs in fractional phase-locked

loops with VCO 727
R. Vishnu, S.S. Anulal & J. Ravi
Adaptive neuro-fuzzy based base station and relay station deployment

for next generation wireless communication 733
M. Vincent
International Conference on Emerging Trends in Computer

Science (ICETICS)
A survey on sentence similarity based on multiple features 743
R.K. Fathima & C. Raseek
A review of extractive sentence compression 751

S.A. Fathima & C. Raseek
A survey of figurative language detection in social media 757
P.D. Manjusha & C. Raseek
A survey of various approaches for interpersonal relationship extraction 765
M. Nijila & M.T. Kala
A survey of morphosyntactic lexicon generation 773
M. Rahul & S. Shine
Classification of question answering systems: A survey 779
S. Sandhini & R. Binu
A survey of lexical simplification 785
K.S. Silpa & M. Irshad
Fast and efficient kernel machines using random kitchen sink and
ensemble methods 793
P. Melitt Akhil & K. Rahamathulla
HOG feature-based recognition for Malayalam handwritten characters 799
E.P. Anjali, A. James & S. Chandran
A novel approach for the veracity and impact prediction of rumors 805
A.R.G. Punneliparambil & N.D. Bisna
A framework for efficient object classification for images having noise
and haze using deep learning technique 811
A. Bhavyalakshmi & M. Jayasree
Bilingual handwritten numeral recognition using convolutional neural network 817
J. Joy & M. Jayasree
xi

A security framework to safeguard the hybrid model of SDN from malicious applications 825
C.M. Mansoor & K.V. Manoj Kumar
A human-intervened CAPTCHA (HI-CAPTCHA) for wireless LAN-based
classroom applications 829
P. Giri, Paras & R. Singh
Semantic identification and representation of Malayalam sentence using SVM 835
S. Bhaskar, T.M. Thasleema & R. Rajesh
A survey on relation extraction methodologies from unstructured text 845
S.K. Bhaskaran & P.C. Rafeeque
Decoupling control of TRMS based on a Relative Gain Array (RGA)
and Kharitonov theorem 851
S.K. Pandey, J. Dey & S. Banerjee
Root ORB—an improved algorithm for face recognition 857
A. Vinay, A. Bharadwaj, A. Srinivasan, K.N. Balasubramanya Murthy & S. Natarajan
Face recognition using SURF and delaunay triangulation 865
A. Vinay, A. Gupta, H. Garg, S. Bhat, K.N. Balasubramanya Murthy & S. Natarajan
Opportunities and challenges in software defined networking and network
function virtualization 873
A. Roy, V.K. Asna, N. Nimisha & C.N. Sminesh
A parallel framework for maximal clique enumeration 879
T. Anusree & K. Rahamathulla
Retrieving and ranking similar questions and data-driven answer selection in
community question answering systems 883
V. Dhrisya & K.S. Vipin Kumar
Convolutional Neural Network (CNN) framework proposed for Malayalam
handwritten character recognition system using AlexNet 889
J. Manjusha, A. James & S. Chandran
A novel technique for script identification in trilingual optical character recognition 895
A. Rone Maria & K.J. Helen
An optimal controller placement strategy using exemplar-based clustering
approach for software defined networking 901
A.G. Sreejish & C.N. Sminesh
Identifying peer groups in a locality based on Twitter analysis 907
K. Sreeshma & K.P. Swaraj
Route choice analysis of metros using smart card data 913
S. Sujix
International Conference on Changing Cities-Architecture and Energy

Management (ICCC)
Architectural applications
Ecological conservation of sites through responsive design: Case of the
Muziris heritage interpretation center 925
M. Gopalakrishnan & S. Surya
xii

Design strategies for daylighting in tropical high rises 933
N. Nabeel, A. John & A.K. Raseena
Internalizing the concept of sustainability—redefining the curriculum of
building construction in architecture schools 943
O.P. Bawane
Policy and assessment scenario

Effectiveness of housing schemes in rural India: A case study of Vellanad, Kerala 951
N. Vijaya, P. Prabhakaran & A. Lakshmi
Planning for solid waste management in Kuttanad wetland region, Kerala State 957
A. Sanil Kumar & V. Devadas
Green infrastructure as a tool for urban flood management 965
R. Suresh & C.A. Biju
Accessibility and interactions with urban blue spaces: A case of Conolly Canal 973
S. Usman Abdulla & C.A. Bindu
Energy efficiency governance in an Indian context 979
V.P. Shalimol & K.M. Sujith
Environmental management of a blue-green network: A case of Valapad
in Thrissur, Kerala 987
N. Basheer & C.A. Bindu
Urban metabolism and food security: Emergy as a metrics link connecting the food
security with the urban spatial aspects for enhanced livability and sustainability 995
B. Narayan & J. Jayakumar
Smart city initiatives

Smart-cities and smart-villages in the Indian context: Some behavioral aspects 1003
P. Varghese
Environment management through meditation: A sustainable approach 1011
M. Yadav
Bicycles for green mobility in urban areas 1019
M. Harisankar & C.A. Biju
Impact of foreign direct investment on the city: Form and growth of cities 1027
A. Mathew & M. Kumar Kini
Author index 1035
xiii

Preface
The 5th Biennial International Conference on ‘Emerging Trends in Engineering Science and
Technology’, ICETEST 2018, was held during 18th to 20th January 2018 at Government
Engineering College, Trichur, Kerala, India. ICETEST is an international interdisciplinary
conference covering research and developments in different domains of Engineering, Archi-
tecture and Management. The distinctive feature of ICETEST 2018 is that it is a culmination
of seven sub conferences; CEASIDE for Civil Engineering, PDME for Mechanical Engi-
neering, PICC for Electrical Engineering, ICAChE for Chemical Engineering, E-SPACE for
Electronics and Communication Engineering, ICETICS for Computer Science Engineering
and ICCC for School of Architecture. In spite of having seven different sub conferences,
there had been an underlying specific common theme – Society, Energy & Environment. Of
late, the energy and environment concern has reached its peak, and researchers in academia
need to support the industry and society through socially and environmentally sustainable
solutions.
ICETEST 2018 was organized at Government Engineering College Thrissur (GECT) to
discuss and disseminate the major advances in technologies for Society, Energy and Environ-
ment. Government Engineering College Trichur, (GECT) established in 1958, is one of the
premier institutions for quality technical education in the state of Kerala, India. GECT offers
undergraduate and post graduate programmes in seven Engineering disciplines and Architec-
ture. Further, GECT is a major research center for four branches of Engineering under the
University of Calicut, and is an approved centre for research under the Quality Improvement
Programme of Ministry for Human Resource Development, Govt. of India. The fifth edition
of biennial international conference ICETEST 2018 is organized as part of Diamond Jubilee
celebrations of the institute.
More than 480 pre-registered authors submitted their manuscripts in the conference under
seven different sub conferences. ICETEST 2018 accepted 258 papers after double blind peer
review process. Finally, 251 authors presented their works in the conference. There were dele-
gates and contributions from countries like, USA, Canada, Korea, Malaysia, UAE and Africa.
Cultural and entertainment programs were arranged along with Gala dinner for the del-
egates on second day evening. Cultural show by world famous Indian institute of Art and
Culture, the ‘Kerala kalamandalam’ was the highlight, which provided a glimpse of Kerala’s
rich cultural heritage. These programs in fact made the technical event richer.
The primary objectives of ICETEST 2018 was to bring the emerging technologies from
various disciplines of Engineering, Science & Technology under the theme Society, Energy
and Environment for the betterment of society and preserving nature. The conference has
provided a platform for showcasing and interacting researchers from different domains.
There were 51 paper presentation sessions in different themes under seven sub conferences,
which were conducted in 13 venues within the institution, with 323 delegates and 748 authors.
For, international delegates, it provided an opportunity to visit Kerala, a place commonly
known as ‘God’s own country’ apart from enjoying the technical fiesta.
xv

Plenary sessions and keynote speeches
There were three plenary sessions and an Editorial and Publishing Workshop at the main
venue.
1. ‘Power Electronics for Renewable Energy and Power Systems: Opportunities and Chal-
lenges’ by Dr. Jian Sun, Professor, Department of Electrical, Computer, and Systems
Engineering (ECSE), Rensselaer Polytechnic Institute, USA.
2. ‘Biomass for Fuels and Chemicals’ by Dr. Suchithra T.G., University of Nottingham,
Malaysia.
3. ‘Power electronics application in alternate energy systems’, by Prof. Ashoka K.S. Bhat,
Professor, Department of Electrical Engineering, University of Victoria, B.C., Canada.
4. ‘Editorial Workshop’ by Dr. Gagandeep Singh Editorial manager for the CRC Press,
Taylor & Francis Group.
Apart from plenary, there were 19 keynote speeches organized by seven constituent sub con-
ferences, with experts from India and abroad. Wide range topics that the keynote sessions
covered and highlighted engross the developments from different domains and challenges
with reference to the theme.
We thank all the distinguished delegates who participated in ICETEST 2018 especially:
• Padma Bhushan Dr. K. Radhakrishnan, Honorary Distinguished Advisor, Department of
Space/ISRO, Govt of India for inaugurating the conference and for consenting to be the
chief guest of the inaugural function.
• Padmasree M. Chandradathan, Scientific Advisor to Chief Minister, Govt. of Kerala, for
consenting to be the chief guest of the valedictory function.
• Renowned publication houses, CRC press, Taylor and Francis and IEEE for their com-
munication and sponsorship.
• The funding agencies TEQIP, AICTE, Govt. of India, DTE and KSCSTE, Govt. of Ker-
ala for the financial support which materialized this event.
• Dr. K.P. Indiradevi, Director of Technical Education for the state of Kerala, for the invalu-
able support and directions.
• Dr. B. Jayanand, the Principal GECT, and Diamond Jubilee conveners, Dr. N. Sajikumar,
Head of Civil Engineering Dr. C.P. Sunilkumar, Professor and Dean UG studies, Dr. M.
Nandakumar, Head of Electrical Engineering, Dr. Thajudin Ahamed V.I., Head of Elec-
tronics and communication Engineering and Dr. V.P. Mohandas, Professor in Mechanical
Engineering, for their valuable guidance and support.
• The members of organising committee Dr. E.A. Subaida, Dept. of Civil Engg. (CEA-
SIDE), Dr. Sudheesh R.S., Dept. of Mechanical Engg. (PDME), Dr. Jaison Mathew, Dept.
of Electrical Engg. (PICC), Dr. Subin Poulose, Dept. of Chemical Engg., (ICAChE), Prof.
Mohanan K.P., Dept. of Electronics & Communication Engg. (E-SPACE), Dr. Swaraj
K.P., Dept. of Computer Science & Engg. (ICETICS) and Dr. Ranjini Bhattathiripad
T., School of Architecture (ICCC) for their help, support, commitment and participation
before, during and after the conference.
• All the conveners and members of various committees, reviewers, volunteers, Faculty,
Staff and Students of GECT for their efforts in materialising the event, ICETEST 2018.
xvii

Conference secretariat and committees
CHIEF PATRONS
Prof. C. Raveendranath—Hon’ble Minister for Education, Govt. of Kerala

Dr. Usha Titus, IAS—Secretary, Higher Education Department, Govt. of Kerala
PATRONS
Dr. K.P. Indiradevi—Director of Technical Education, Govt. of Kerala

Dr. B. Jayanand—Principal, Govt. Engineering College Trichur
ORGANISING SECRETARY
Dr. Jiji K.S.—Associate Professor, Electrical Engineering, GEC TCR

Dr. Rajesh Vanchipura—Associate Professor, Mechanical Engineering, GEC TCR
JOINT SECRETARY
Dr. Anjana R.—Asst. Professor, Chemical Engineering, GEC TCR
TREASURER
Prof. K.R. Rahul—Asst. Professor, Mechanical Engineering, GEC TCR
ORGANIZING COMMITTEE
Dr. E.A. Subaida—Dept. of Civil Engg.

Dr. Sudheesh R.S.—Dept. of Mechanical Engg.
Dr. Jaison Mathew—Dept. of Electrical Engg.
Dr. Subin Poulose—Dept. of Chemical Engg.
Prof. Mohanan K.P.—Dept. of Electronics & Communication Engg.
Dr. Swaraj K.P.—Dept. of Computer Science & Engg.
Dr. Ranjini Bhattathiripad T.—School of Architecture.
INTERNATIONAL ADVISORY COMMITTEE
Dr. Lars-Erik Lindgren—Professor, Department of Engineering Sciences and Mathematics,

Division of Mechanics of Solid Materials, Lulea University of Technology, Sweden
xix

Dr. Angappa “Guna” Gunasekaran—Dean and Professor, Charlton College of Business,
University of Massachusetts, Dartmouth, USA
Dr. Krishna K. Krishnan—Professor and Chair, Department of Industrial, Systems and
Manufacturing Engineering, Wichita State University, Kansas
Dr. Chandramouli Viswanathan Chandramouli—Associate Professor of Civil Engineering,
College of Engineering and Sciences, Hammond
Dr. Annick Anctil—Assistant Professor, Civil and Env Engg., Michigan State University, USA
Dr. J.S. Vinod—Senior Lecturer in Civil Engineering, Faculty of Engineering and Information
Sciences, University of Wollongong, NSW 2522, Australia
Dr. Antony Joseph—Assistant Research Scientist, Energy and Building Research Center
(EBRC), Kuwait Institute for Scientific Research (KISR), Kuwait
Prof. Seung-Ki Sul—Seoul National University, South Korea
Prof. Frede Blaabjerg—Aalborg University, Denmark
Dr. Tomy Sebastian—Halla Mechatronics, USA
Prof. Adelino Jorge Lopes Ferreira—Director of the Road Pavements Laboratory, Professor,
Department of Civil Engineering, University of Coimbra, Portugal
Prof. P.A. Muhammed Basheer, FREng—Professor of Structural Engineering, Head of
School of Civil Engineering, University of Leeds, West Yorkshire, UK
Dr. Srinivas Sriramula—Lloyd’s Register Foundation (LRF), Centre for Safety & Reliability Engi-
neering, School of Engineering Fraser Noble Building, University of Aberdeen, UK
Dr Vivek Balachandran—Singapore Institute of Technology
NATIONAL ADVISORY COMMITTEE
Dr. Devdas Menon—Professor in Structural Engineering, IIT, Madras

Dr. K.P. Sudheer—Professor in Civil Engineering, IIT, Madras
Dr. K. Vijayakumar—Director of Technical Education (Rtd.), Kerala
Dr. Kuncheria P. Isaac—Vice Chancellor, Kerala Technological University
Dr. J. Letha—Vice Chancellor, Cochin University of Science and Technology, Kerala
Dr. Suresh Nair—Chair, IEEE, Kerala
Prof. (Dr.) N. Kavita Daryani Rao—Vice Chancellor, Jawaharlal Nehru Architecture and Fine
Arts University, Telangana
Prof. (Dr.) Vasudha Gokhale—Professor in Archtecture, B.N.C.A College of Architecture for
Women, Maharashtra
Dr. Balaji C.—Professor, Dept. of Mechanical Engineering, IIT, Madras
Er. A.F. Antony—IDSE Director, Military Engineer Service, CE(AF), Bangalore
Dr. P.V. Venkitakrishnan—Director, ISRO Propulsion Complex, Mahendragiri, Tamil Nadu
Er. Shajan T.I.—Group Director-GTRE, DRDO, Bangalore
Dr. Krishna Vasudevan—Professor, Dept. of Electrical Engg., IIT, Madras
Dr. K. Gopakumar—Professor, Dept. of Electrical Engg., IISc, Bangalore
Prof. A.K. Nandakumaran—Department of Mathematics, IISc, Banaglore
Dr. Hari M. Varma—Department of Bio-sciences and Bio-engineering IIT, Bombay
Dr. Subrahmanyam Kalyanasundaram—Assistant Professor, Dept. of Computer Science and
Engineering, IIT, Hyderabad
Dr. R. Anandalakshmi—Dept. of Chemical Engg., IIT, Guwahati
Dr. G. Pugazhenthi—Professor, Department of Chemical Engineering, IIT, Guwahati
Dr. Sabu Sebastian—Scientist F, NPOL, Kochi
Prof. (Dr.) Anandhakuttan B. Unnithan—IIM, Kozhikode
Mr. Bhav Singh—Scientist F, DMRL, Hyderabad
Dr. Vinod S.—Asst. Professor, NIT, Trichy
Dr. Prasad Krishna—Professor, NIT, Surathkal
Dr. Mohammad Idrees—Professor & Chairman, Department of Chemical Engineering, Zakir
Husain College of Engineering & Technology, Aligarh Muslim University, Aligarh
Dr. A. Unnikrishnan—Former Outstanding Scientist, NPOL, Kochi
xx

TECHNICAL COMMITTEE
Dr. R. Rudramoorthy—Principal, PSG College of Technology, Coimbatore

Dr. N. Ramachandran—Professor and Head, Dept. of Mech. Engg., NIT, Calicut
Dr. Jyothi Prakash—Professor in Civil Engineering, IIT, Bombay
Dr. M.V.L.R. Anjaneyulu—Professsor in Civil Engineering, NIT, Calicut
Dr. Chandrakaran S.—Professsor in Civil Engineering, NIT, Calicut
Dr. N. Ganesan—Professsor in Civil Engineering, NIT, Calicut
Mr. Suthakar S.—Asst. Manager, BEML, Mysore
Dr. Kamlesh Kumari—Professor in Chemical Engineering, SLIET, Punjab
Dr Pushpa Jha—Professor, Dept. of Chemical Engg., SLIET, Punjab
Dr. Lity Alen Varghese—Department of Chemical Engineering, NIT, Calicut
Dr. Shiny Joseph—Department of Chemical Engineering, NIT, Calicut
Dr. A. Sreenivasan—Department of Biotechnology, NIT, Raipur
Dr. Saravanan Chandran—Assistant Professor, NIT, Durgapur
Dr. Sudeep K.S.—Assistant Professor, NIT, Calicut
Mr. Muralidharan—DGM, MIL Controls
Mr. Unnikrishnan P.K.—Head, Apollo Tyres, Kerala
xxi

Editors
Dr. Rajesh Vanchipura—Associate Professor, Department of Mechanical Engineering,

Government Engineering College Trichur, Thrissur, Kerala, India. 680 009, rajeshvanchipura@
gmail.com, Phone: +91 944739505
Dr. Jiji K.S.—Associate Professor, Department of Electrical Engineering, Government
Engineering College Trichur, Thrissur, Kerala, India. 680 009, [email protected], Phone: +91
9446897218
EDITORS FOR SIX SECTIONS
Section 1: Ceaside; Civil Engineering Advancements for Sustainable Infrastructure

Development and Environment
Editor: Dr. E.A. Subaida—Associate Professor, Department of Civil Engineering, Government
Engineering College Trichur, Thrissur, Kerala, India. [email protected]
Section 2: PDME; Progressive Developments in Mechanical Engineering
Editor: Dr. R.S. Sudheesh—Associate Professor, Department of Mechanical Engineering,
Government Engineering College Trichur, Thrissur, Kerala, India. [email protected]
Section 3: ICAChE; International Conference on Advances in Chemical Engineering
Editor: Dr. SubinPoulose—Associate Professor, Department of Chemical Engineering,
Section 4: E-SPACE; Electronics, Signal Processing and Communication Engineering
Editor: Prof. Mohanan K.P.—Associate Professor, Department of Electronics and Com-
munication Engineering, Government Engineering College Trichur, Thrissur, Kerala, India.
[email protected]
Section 5: ICETICS; International Conference on Emerging Trends in Computer Science
Editor: Dr. Swaraj K.P.—Associate Professor, Department of Computer Science and Engineering,
Section 6: ICCC; International Conference on Changing Cities- Green Architecture and
Energy Management
Editor: Dr. Ranjini Bhattathiripad T.—Professor, School of Architecture, Government
Engineering College Trichur, Thrissur, Kerala, [email protected]
xxiii

Civil Engineering Advancements for Sustainable
Infrastructure Development and Environment
(CEASIDE)
Structural mechanics and materials
ICETEST 2018_Book.indb 1 7/9/2018 12:12:25 PM

Studies on ambient cured geopolymer concrete
Amy De Sam & Raj S. Deepa

College of Engineering, Trivandrum, Thiruvananthapuram, Kerala, India
ABSTRACT: Cement is the main constituent present in the production of concrete. The
production of Ordinary Portland Cement (OPC) leads to huge emission of green house gases
such as CO2. Fly ash based geopolymer binders are an innovative alternative to OPC which
can provide high strength. Much research has been conducted on the properties of fly ash
based geopolymer concrete hardened by heat curing, which is considered as a limitation for
cast in situ applications of geopolymers. The aim of this study is to generate geopolymers
cured without elevated heat. For this Ground Granulated Blast Furnace Slag (GGBFS) is
also used as a binder along with fly ash in different percentages. The result obtained revealed
that the addition of GGBFS in fly ash based geopolymer concrete enhanced its mechanical
properties in an ambient curing condition.
Keywords: fluid–Geopolymer concrete, fly ash, blast furnace slag, ambient curing
1 INTRODUCTION
Concrete is a widely used construction material all over the world. To reduce the environmental
impact caused by the production of cement, it is necessary to develop some environmentally
friendly alternative materials. For this a material called geopolymer, synthesized by alkali
activation of aluminosilicate compounds can be used. Geopolymer is a binder material which
can be a sustainable and economical binding material as it is produced from industrial by-
products such as fly ash replacing 100% of cement in concrete. As compared with Portland
cement, geopolymers have very low CO2 emissions (Partha Sarathi et al. 2014).
Low calcium fly ash (class F) is used as a suitable material for geopolymer because of its
wide availability and lower water demand. From previous investigations it is clear that heat
cured low calcium fly ash based geopolymer concrete (GPC) has shown excellent mechani-
cal and durability properties (Karthik A. et al. 2017). The only limitation of fly ash based
GPC is that, it requires heat curing, so this cannot be used for cast in-situ applications at low
ambient temperature.
To widen the use of GPC beyond precast applications it is necessary to produce GPC mem-
bers cured in ambient conditions. For this Ground Granulated Blast Furnace Slag (GGBFS)
can be used as a binder along with fly ash (Pradip Nath et al. 2014). GGBFS is a by-product
from the steel industry with a high calcium content compared to fly ash. Blast furnace slag
is formed during the production of hot metals in blast furnaces. If the molten slag is cooled
and solidified by rapid quenching, GGBFS is formed. Only a few research papers are avail-
able regarding slag based geopolymer concrete, and do not provide any clear evaluation of
data. Thus to establish the relevance of GGBFS in GPC it is necessary to thoroughly study
its mechanical properties.
2 EXPERIMENTAL PROGRAM
The experiment program consists of developing slag based GPC by replacing the amount
of fly ash in GPC by various percentage such as 10%, 20%, 30% and 40%. To study the

fresh properties, a slump test was conducted. For assessing the mechanical properties, a
compressive strength test, split tensile strength test and flexural strength test were performed.
2.1 Material properties

Class F fly ash containing 21.4% calcium oxide and 64.84% silica (SiO2) having an average
particle size of less than 45µ and GGBFS containing 38.34% calcium oxide and 34.26% silica
(SiO2) is used as binder materials. The fine aggregate used is M-sand. The fineness modulus
of M-sand is 2.94 conforming to grading zone II. The coarse aggregate used for this study is
crushed granite of 20 mm nominal size having specific gravity of 2.7. The alkaline activator
used is sodium silicate (solution) and sodium hydroxide (solution). The ratio of alkaline acti-
vator to binder is taken as 35% and the ratio of sodium silicate to sodium hydroxide is taken
as 2.5% [Pradip Nath, (2014)]. A superplasticizer is used to adjust the workability of mixes.
2.2 Mix design

Since no standard procedure is available for the mix design of GPC, the mix proportion for the
geopolymer concrete is adopted from the study conducted by Ganesan et al., (2015) [Ganesan,
N., (2015)]. For the preparation of slag based geopolymer concrete, the mix proportion was
modified by replacing fly ash with 10% to 40% of GGBFS, while the other ingredients remained
the same. The designations used for identifying the mixes are M-mix having no GGBFS and
100% fly ash, M1-mix having 10% GGBFS and 90% fly ash, M2-mix having 20% GGBFS
and 80% fly ash, M3-mix having 30% GGBFS and 70% fly ash, M4-mix having 40% GGBFS
and 60% fly ash respectively. The mix proportion of all the mixes is shown in Table 1.
2.3 Preparation of specimens

Firstly coarse aggregate in the saturated surface dry condition was dry mixed with fly ash,
GGBFS and sand in a laboratory mixer for three minutes. To produce geopolymer concrete
alkaline activator solution, superplasticizer and extra water was added to the dry mix and the
mixing continued for five minutes. Immediately after mixing, the slump of fresh concrete was
measured. Standard cubes of size 150 mm, cylinders of 150 mm diameter and 300 mm height
and prisms of size 100 mm × 100 mm × 500 mm were prepared for determining the hardened
properties. The moulds were filled in layers and each layer was well compacted and kept in
ambient conditions. After 24 hours of casting the specimens were removed from the moulds
and allowed to cure under ambient conditions for seven days.
2.4 Testing
The workability of fresh geopolymer concrete was determined by a slump test. After seven
days of casting, a compressive strength test, splitting tensile strength test and flexural strength
test were carried out using a compression testing machine with a capacity of 3000 kN.
Table 1. Concrete mix proportions in kg/m3.
Mix M M1 M2 M3 M4
CA 1248 1248 1248 1248 1248

FA 600 600 600 600 600
Fly ash 408 367.2 326.4 285.6 244.8
GGBFS 0 40.8 81.6 122.4 163.2
Na2SiO3 Soln. 103 103 103 103 103
NaOH Soln. 41 41 41 41 41
Superplasticizer 10.2 10.2 10.2 10.2 10.2
Water 14.5 14.5 14.5 14.5 14.5

3 RESULTS AND DISCUSSION
3.1 Slump and compressive strength

Slump values of fresh geopolymer concrete and compressive strengths of specimens following
seven days of ambient curing are shown in Table 2.
From Table 2, it can be seen that the workability of the mix reduced with the increase
in GGBFS content. This may be due to the irregular shape of slag particles in comparison
with the spherical shape of fly ash since it reduces the ability for free movement. Accelerated
reactions occurred in the slag containing mix and this could also be a reason for reduced
workability. When 40% of GGBFS is added to GPC, the workability reduces to a very low
value, therefore it is difficult to work with this in practice.
From Table 2, it can be noted that the addition of GGBFS increased the strength prop-
erties. This increase was mainly due to the production of a calcium based compound in
the dissolved binder which produces a reaction product from both the slag and fly ash.
From the compressive strength test results shown in Table 2, it is seen that the compressive
strength increases up to 30% of GGBFS content. This increase in compressive strength by
the addition of GGBFS is due to its pore filling effect [2]. However, when the percentage of
GGBFS is increased to 40%, the strength is reduced, this may be due to the very low work-
ability of the mix and a reduction of the geopolymerization process due to the increase in
calcium content.
3.2 Splitting tensile strength

Tensile strength of concrete is an important mechanical property related to the design aspects
of concrete structures, since it relates to crack initiation and propagation, as well as shear and
anchorage of reinforcing steel in concrete. The splitting tensile test is usually conducted for
determining the tensile strength of concrete. The results of seven days of the splitting tensile
strength test are shown in Table 3.
From Table 3 it can be seen that split tensile strength increases as the amount of GGBFS
increases from 10% to 30%. The addition of an extra amount of water in the mixtures will
affect the rate of tensile strength development of geopolymer concrete [3]. For this reason the
tensile strength of the M4 mix is reduced.
3.3 Flexural strength

The flexural strength values of specimens at the age of seven days are shown in Table 4.
From Table 4, it can be seen that the flexural strength is increasing upto 30% of GGBFS
and then decreases when the percentage of GGBFS is increased to 40%. Figure 1 shows the
percentage increase in strength with different percentages of GGBFS.
From Figure 1, it can be concluded that the addition of 30% GGBFS gives higher strengths
so this can be taken as the optimum percentage.
Table 2. Slump and compressive strength.

Table 3. Splitting tensile strength in N/mm2.
Compressive
% of Slump strength Split tensile %
Mix GGBFS (mm) (N/mm2) Mix strength variation
M – 125 23.5 M9 1.33 0

M1 10 120 25 M1 1.38 3.76
M2 20 110 28 M2 1.42 6.77
M3 30 95 48.8 M3 3.112 133.8
M4 40 20 31 M4 2.82 112

Table 4. Flexural strength in N/mm2.
Flexural %
Mix strength variation
M 4.3 0
M1 4.8 11.63
M2 5.4 25.58
M3 7.8 81.4
M4 7.2 67.44
Figure 1. Percentage increase in strengths.
4 CONCLUSIONS
From the study conducted the following conclusions can be drawn:-

1. The workability of GPC mix decreases with an increase in GGBFS content. The work-
ability of GPC containing 40% GGBFS is reduced by 84% compared to that containing
no GGBFS.
2. The mechanical properties such as compressive strength, split tensile strength, and flexural
strength increases with an increase in GGBFS content.
3. With 30% of GGBFS in the mix better workability, compressive strength, split tensile
strength and flexural strength were obtained.
4. It can be concluded that 30% GGBFS content can be taken as the optimum percentage of
GGBFS in geopolymer concrete based on fresh and hardened properties.
REFERENCES
Aradhana Mehta, Kuldeep Kumar (2016). Strength and durability characteristics of fly ash and slag
based geopolymer concrete, International Journal of Civil Engineering & Technology (IJCIET)
Research, 305–314.
Ganesan, N., Ruby Abraham & Deepa Raj S. (2015). Durability charecteristics of steel fibre reinforced
geopolymer concrete, Construction and Building Materials, 93, 471–476.
Karthik A., Sudalaimani K. & Vijaya Kumar C.T. (2017). Investigation on mechanical properties of fly
ash-ground granulated blast furnace slag based self curing bio-geopolymer concrete, Construction
and Building Materials, 149, 338–349.
Partha Sarathi, Pradip Nath, Prabir Kumar Sarker. (2014). The effect of ground granulated blast-furnace
slag blending with fly ash and activator content on the workability and strength of geopolymer
concrete cured at ambient temperature, Materials and Design, 62, 32–39.
Phoo-Ngernkham, T., Chindaprasirt, P., Sata, V. & Sinsiri, T. (2013). High calcium fly ash geopolymer
containing diatomite as additive, Indian Journal of Engineering Material Science, 20(4), 310–318.
Pouhet Raphaelle, Cyr Martin. (2016). “Formulation and performance of fly ash metakaolin geopolymer
concretes, Construction and Building Materials, 120, 150–160.
Pradip Nath, Prabir Kumar Sarkar. (2014). Effect of GGBFS on setting, workability, and early strength
properties of fly ash geopolymer concrete cured in ambient condition, Construction and Building
Materials, 66, 163–171.
Rashad, A.M. (2013). A comprehensive overview about the influence of different additives on the prop-
erties of alkali activated slag: A guide for civil engineer, Construction and Building Materials, 47,
29–55.

Relationship between the modulus of elasticity and characteristic

compressive strength of B. subtilis impregnated concrete
C.M. Meera & V. Subha

School of Engineering, Cochin University of Science and Technology, Kerala, India
ABSTRACT: The modulus of elasticity is a very important property of concrete as a material

subjected to loading and displacement. Compressive strength shall be considered as an indica-
tor of the modulus of elasticity of concrete. This study focused on experimentally achieving the
relationship between these two parameters. Conventional concrete and Bacillus subtilis bacteria
impregnated concrete were considered for the investigation. The elastic properties of conven-
tional and bacterial concrete with various grades of concrete namely M15, M20 and M25 were
used. The study resulted in arriving at a governing equation to show the relationship between
conventional and bacterial concrete. The validity of the equation was also investigated.
Keywords: Bacillus subtilis, compressive strength, modulus of elasticity
1 INTRODUCTION
The modulus of elasticity shows the influence of material strain subjected to loading and
deflection due to loading. It is a very important parameter of a material to predict its struc-
tural behavior. It is a very important mechanical property of concrete. As the modulus of
elasticity increases, the material becomes stiffer and the deflection experienced by the struc-
ture decreases. The modulus of elasticity of concrete shall be calculated from a compressive
strength test. Stresses and strains are obtained from the strength tests and plotted. The elastic
modulus is then measured as the slope of the curve at 40% of the stress value at the ultimate
load. It is necessary to obtain the value of the modulus of elasticity to perform simulation of
the material behavior. The modulus of elasticity and compressive strength possess a relation-
ship, and both increase with the progress of the hydration process of cement in the case of
concrete. Hence, it is important to monitor the parameter.
Concrete is the most widely used construction material all over the world. Concrete struc-
tures are often subject to cracks which lead to durability problems for such structures. Cracks
can develop at any stage of the service life of concrete due to shrinkage, loading, weathering
and so on. The cracks not only permit the entry of corrosive fluids but also aid in further
propagation and damage the structure. Micro-cracks in concrete could allow entry of water
and other impurities like chloride and sulfate ions into the concrete, which lead to degra-
dation of the concrete matrix and corrosion of embedded reinforcement. This affects the
structural integrity of concrete. The expenses incurred in maintenance and repair of concrete
structures subject to cracks are very high. Many techniques are being used to arrest and
repair these cracks to enhance the durability of such structures. One such technique is bio-
engineered concrete which is intrinsic in nature. This could be developed by adding a special
type of calcite precipitating bacteria to concrete. The bio-mineralization capacity of bacteria
is utilized in this technique to fill the cracks. By conducting feasibility studies of different
aspects, various researchers have concluded that self-healing behavior can be achieved by
impregnating calcite precipitating bacteria into concrete (Jonkers & Schalengon, 2009).
It is important to analyze the relationship between the modulus of elasticity and compres-
sive strength of bacterial concrete and compare it with that of conventional concrete for the

further use of bacterial concrete as a construction material. Therefore, the current study was
conducted to shed light on the elastic behavior of bacterial concrete when compared with
that of normal concrete.
2 LITERATURE REVIEW
Bacterial concrete is emerging as a smart material with wholesome potential applications in

civil infrastructure. It can be a sustainable solution to durability problems in concrete due
to its intrinsic healing behavior and capability to restore the original functionality of the
structure. In bacterial concrete, when water enter the cracks, the bacterial action takes place
in the presence of calcium based carrier materials to produce calcite crystals to fill the cracks,
thereby preventing further entry of water (Wang et al., 2010). Bacterial concrete, as a self-
healing system, can accomplish a substantial cost reduction in terms of health monitoring,
damage detection and maintenance of concrete structures, assuring a safe service life of the
structure (Meera & Subha, 2016). Such a system, if developed in a practicable technique,
can become a breakthrough in the construction industry. The feasibility of this system of
bacterial concrete as a self-healing method under various aspects are being inspected.
The modulus of elasticity is a very suggestive mechanical parameter reflecting the ability of
the concrete to deform significantly (Krystian & Stefania, 2015). Elasticity is a way to exemplify
the mechanical response of the material form for the applied stresses that occur within the lin-
ear regime. By knowing the elastic behavior of the material, it could be assessed for its applica-
bility. Cement content, aggregate volume fraction, and fiber content and compressive strength
have a strong correlation with the elastic properties of concrete (Sedat et al., 2013).
The development of a self-healing system in a practicable technique, could become a
breakthrough in the construction industry. Research based purely on self-healing bacterial
concrete is rare. The feasibility of the bacterial concrete as a self-healing method under vari-
ous aspects should be inspected. A study was carried out to find a relationship with the grade
of concrete and modulus of elasticity of the bacterial concrete.
3 MATERIALS AND METHODS
3.1 Materials
The materials used in this study include the following:
3.1.1 Cement
Ordinary Portland Cement (53 grade), which has been tested for various properties as per
IS:4031-1988 and found to be conforming to various specifications of IS:12269-2013 was used.
3.1.2 Fine aggregates

M-sand with a specific gravity of 2.69, conforming to IS:383 zone II and passing through a
4.75 mm sieve was used as fine aggregate. The sand was tested as per IS:2386 (Part III)-1963.
3.1.3 Coarse aggregates

Coarse aggregates of granite metal of 20 mm, having a specific gravity of 2.77 and a fineness
modulus of 6.9%, tested as per IS:2386-1963 was used.
3.1.4 Water
Locally available potable waterconforming to standards specified in IS:456-2000 was used for
mixing and curing.
3.1.5 Microorganism
Bacillus subtilis (B. subtilis), a laboratory cultured bacterium collected from Kerala
Agricultural University Mannuthy, was used. The bacteria, in suspension with a concentration
8

Table 1. Mix proportion of concrete specimens.
Water–
Concrete Fine Coarse cement
grade Cement aggregates aggregates ratio
M15 1 1.76 3.71 0.45

M20 1 1.85 3.13 0.5
M25 1 1.86 3.12 0.4
of 108 cells/ml, was collected and an optimum concentration of 105 cells/ml bacteria was
obtained from the sample.
3.2 Mix proportion

M15, M20 and M25 grades of concrete were adopted for testing. The mix proportions of the
materials used in the concrete are listed in Table 1. For bacterial concrete, microorganisms
were added in suspension in mixing water and the same proportions were adopted.
3.3 Specimens
Test specimens consisted of standard cube specimens of side 150 mm and standard cylinder
specimens of dimensions 150 mm diameter and 300 mm height.
3.4 Strength assessment

The cubes and cylinders were prepared for concrete mixes of M15, M20 and M25 grades
of concrete with and without the addition of B. subtilis. Cubes were tested for compressive
strength at seven days and 28 days. Cylinders were used for examining elastic moduli at
28 days of curing.
3.5 R
elationship between Young’s modulus and characteristic compressive strength
of bacterial concrete
The tests to find the elastic modulus and compressive strength of concrete were conducted
in a compression testing machine. To determine the elastic modulus, a compressometer was
attached to the test cylinder. A zero reading was affirmed before exposing the cylinder to
uniaxial compression. For an interval of every 20 kN load, deflection values on the cylinder
were noted. A stress–strain graph was plotted for the load and deflection values. Then, the
elastic modulus was determined by the slope of the line drawn to 40% of the stress value at
the ultimate load. The values of the elastic modulus corresponding to the characteristic com-
pressive strength were compared by plotting a graph of the same.
4 RESULTS
The compressive strength of bacterial concrete specimens showed a significant increase

of average 28-day compressive strength when compared to that of conventional concrete,
for all the evaluated grades of concrete. The average 28-day compressive strength of bac-
terial concrete of M15 grade of concrete showed an increase of 10% to that of conven-
tional concrete. For M20 and M25 grades of concrete, the average 28-day compressive
strength increased by 11% and 26%, respectively. The graphical representation of average
compressive strength of conventional and bacterial concrete for different grades is shown
in Figure 1. The observed values of the modulus of elasticity of bacterial concrete of all the
assessed grades of concrete, also showed a considerable increase in values when compared
9

to that of conventional concrete without the microorganism. The modulus of elasticity val-
ues of bacterial concrete of grades M15, M20 and M25 were found to be increased by 6%,
4% and 6%, respectively, when compared to that of conventional concrete. Figure 2 shows
a graphical representation of the modulus of elasticity of different grades of conventional
and bacterial concrete.
A relationship between modulus of elasticity and characteristic compressive strength of
bacterial concrete was obtained with a squared regression value of 99.8% with the help of
Microsoft Excel software. The expression of relation is as follows:
E = 600.8fck + 11474 (1)
where,
E is the modulus of elasticity of concrete,
fck is the characteristic compressive strength of concrete.
Figure 1. Test set up for finding the modulus of elasticity of concrete.
Figure 2. Average 28-day compressive strength of conventional and bacterial concrete.
10

Figure 3. Modulus of elasticity of conventional and bacterial concrete.
Figure 4. Relationship between Young’s modulus and characteristic compressive strength of bacterial
concrete and governing equation.
Figure 3 shows a graphical representation of the relationship between the modulus of elas-

ticity and characteristic compressive strength of bacterial concrete, from which the governing
Equation 1 was obtained.
5 VALIDATION
The validation of the equation obtained from the study was conducted with the experimental
investigations as per (Ganesh & Siddiraju, 2010). In their paper, for M40 grade bacterial
concrete, they experimentally obtained a value for the modulus of elasticity of concrete of
34235 N/mm². From the governing equation of the relationship between the modulus of
elasticity and characteristic compressive strength of bacterial concrete as per Equation 1,
the value of the modulus of elasticity for M40 grade of concrete obtained is 35506.4 N/mm2,
which has a percentage variation of only 3.8% from the value observed in the study. Hence,
it could be inferred that the governing equation of the relationship between the modulus of
elasticity and characteristic compressive strength of bacterial concrete holds good.
11

6 CONCLUSION
This experimental study shows that impregnation of B. subtilis into concrete shows
improvements in the compressive strength and modulus of elasticity of concrete. On carrying
out compressive strength tests on M15, M20 and M25 grades of concrete, an average increase
in compressive strength of 15.8% was observed which is a substantial enhancement of the
strength parameter. Investigation of elastic properties of M15, M20 and M25 grades of
concrete showed a sizable enhancement of the modulus of elasticity value of 5.6% on average.
The governing equation of the relationship between Young’s modulus and characteristic
compressive strength of bacterial concrete appears reliable on validation with the reference
literature, with a variation of 3.8% which could indicate that the governing equation holds
good. Results overall show a significant improvement in properties of concrete and evidently
proves bacterial concrete to be a feasible solution, while shedding light on future scope for
use in construction.
REFERENCES
Ganesh B.N., Siddiraju S., (2016). An experimental study on strength and fracture properties of self
healing concrete. International Journal of Civil Engineering and Technology (IJCIET), 7(3), 398–406.
Gavimath C.C., Mali B.M., Hooli V.R., Patil A.B., (2011). Potential application of bacteria to improve
the strength of cement concrete. International Journal of Advanced Biotechnology and Research, 3(1),
541–544.
Jonkers H.M., Arjan Thijssen, Gerad Muyzer, Oguzhan Copuroglu, Erik Schlangen, (2010). Appli-
cation of bacteria as self healing agent for the development of sustainable concrete, Ecological
Engineering, 36(2), 230–235.
Jonkers H.M., Erik Schalengon, (2009). A two composite bacteria based self healing concrete. Concrete
Repair, Rehabilitation and Retrofitting, 2(1), 215–220.
Kim V.T., Elke Gruyert, Hubert Rahier, Nele D.B., (2012). Influence of mix composition on the extent
of autogenous crack healing by continued hydration or calcium carbonate formation. Construction &
Building Materials, 37(1), 349–359.
Kim V.T., Nele D.B., Willem D.M., (2010). Use of bacteria to repair cracks in concrete Cement Concrete
and Research, 40(1), 157–166.
Klaas V.B., (2012). Self-healing material concepts as solution for aging infrastructure. 37th Conference
on Our World in Concrete & Structures, Singapore, 8(1), 29–31.
Krystian J., Stefania G., (2015). The influence of concrete composition on Young’s modulus. Procedia
Engineering, 108(1), 584–591.
Meera C.M., Dr. Subha V., (2016). Strength and durability assessment of bacteria based self-healing
concrete. IOSR Journal of Mechanical and Civil Engineering (IOSR-JMCE), Proceedings of the
International Conference on Emerging Trends in Engineering and Management, 1(1), 1–7.
Sedat K., Leyla T., Halit Y.E., (2013). Young’s modulus of fiber-reinforced and polymer-modified
lightweight concrete composite, Construction & Building Materials, 22(6), 1019–1028.
Song G., Ma N. & Li H.N. (2006). Applications of shape memory alloys in civil structures. Engineering
Structures, 28(19), 1266–1274.
Sonja T., Gan S.N. & Noor H.A.K., (2011). Optimization of microencapsulation process for self-healing
polymeric material. Sains Malaysiana, 40(7), 795–802.
Trask R.S., Williams G.J. & Bond, I.P. (2007), Bioinspired self-healing of advanced composite structures
using hollow glass fibres. Journal of the Royal Society Interface, 4(1), 363–371.
Virginie Wiktor, & Henk M. Jonkers, (2011). Quantification of crack healing in novel bacteria based self
healing concrete. Cement and Concrete Composites, 33(7), 763–770.
Wang J.Y., Tittelboom K.V., Belie N.D., Verstraete W., (2010). Potential of applying bacteria to heal
cracks in concrete. Proceedings of Second International Conference on Sustainable Construction
Materials and Technologies, Universita Politechnica delle Marche, Ancona, Italy, 06(1), 28–30.
12

Effect of copper slag as fine aggregate in high strength concrete
A. Sreelakshmi & G. Jeenu

ABSTRACT: This paper reports an experimental investigation on the effect of copper

slag as partial replacement of sand in high strength concrete. The fine aggregate in the high
strength mix (M60) is partially replaced with 20%, 40% and 60% copper slag by weight.
Rheological properties, mechanical properties such as compressive strength, split tensile
strength, flexural strength and fracture properties of the control mix and the modified mixes
were examined. By the replacement of fines by copper slag, there is improvement in work-
ability characteristics, tensile and flexural strength, and fracture energy and fracture tough-
ness. These properties showed optimal improvement for 40% replacement of conventional
aggregates with copper slag.
1 INTRODUCTION
Construction industry is one of the major contributors towards the fast growing Indian econ-
omy. Concrete is the most utilized material by the construction industry due to its excellent
properties and versatility (Sivakumar & Sivaramakrishnan 2016). The per capita consump-
tion of concrete in India is more than 200 kg. This huge demand for concrete, which is
homogeneous, made of heterogeneous materials such as cement, sand and aggregates, creates
heavy exploitation of natural resources, especially aggregates (Patnaik et al. 2014). The large
consumption of natural aggregates beyond the limit will cause damage to the environment.
Conservation of natural resources and preservation of environment being the essence of any
development, the researchers are in constant pursuit for alternate materials that can replace
these natural materials, especially aggregate. Even if there are artificial aggregates currently in
use, their manufacturing creates lots of ecological issues. Another problem arising from con-
tinuous technological and industrial development is the disposal of waste material. Hence,
the last decade witnessed many researches and innovative solutions on the utilization of
waste products in concrete. One such development was the utilization of industrial wastes
and demolition wastes as aggregates, which provides an alternative to the natural and artifi-
cial aggregates (Al-jabri et al. 2009).
Copper is one of the most widely used metal next to steel and aluminium, and the annual
production is about ten lakh tonnes (Tiwari & Saxena 2016). Copper slag is a by-product
obtained during smelting and refining of copper. This waste product is managed by recov-
ering of metal, production of value added products, recycling and disposal by dumping
(Khanzadi & Behnood 2009). Copper slag, shows in its chemical composition high contents
of aluminium, silica and iron oxides, similar to that of pozzolanic materials. Additionally,
its hardness and gradation indicate its suitability as an alternative material for fine aggre-
gate (Ambily et al. 2015). Hence, copper slag can be used as a replacement material for fine
aggregates in production of concrete. By using this in construction field, the pollution due to
other methods of slag disposal can be minimised. Being a waste material, the material cost is
a minimum and will result as a cost effective solution for aggregates (Tamil selvi et al. 2014).
So the use of copper slag in concrete provides ecological as well as economic benefits.
13

As copper slag is produced in the country in large amounts, its utilization in construction will
solve the problem of environmental pollution, waste disposal and recycling (Wei wu et al. 2009).
Though the use of copper slag in concrete production has attracted considerable attention in
the past years, very few researches has been reported on its effect in high strength concrete. The
effect of fracture parameters of such concretes is still not explored. Hence, it is worth studying the
effect of copper slag as fine aggregate in high strength concretes. This experimental study investi-
gates the possibility of replacing fine aggregates with copper slag in different percentages in high
strength concrete by determining its influence in rheological, mechanical and fracture properties.
2 EXPERIMENTAL PROGRAMME
2.1 Materials
The materials used for the investigation were cement, fine aggregate, coarse aggregate of
12.5 mm nominal size, superplasticizer and water. The constituent materials are tested as per
the methods prescribed by the relevant IS codes.
2.1.1 Cement
Standard tests were conducted as per IS on the 53 grade OPC used for the investigation.
28th day compressive strength obtained was 54 MPa. Other properties obtained as Standard
consistency = 34%, Initial setting time = 130 minutes, Final setting time = 540 minutes. The
results conform to the values specified in IS code 12269:1987.
2.1.2 Coarse aggregate

Crushed stone aggregate of maximum size 12.5 mm was used in the study. Physical properties
were tested according to IS 2386:1963 and the results are obtained as specific gravity = 2.68,
bulk density = 2.65 g/cc, void ratio = 0.67 and porosity = 40%. Gradation curve revealed that
the aggregate conforms to the limits specified by IS 383:1970.
2.1.3 Fine aggregate

2.1.3.1 Crushed stone fine aggregates
Crushed stone aggregates passing through 4.75 mm sieve were used as fine aggregate. The
physical properties of fine aggregate tested as per IS 2386:1963 were obtained as specific
gravity = 2.38, bulk density = 1.74 g/cc, void ratio = 0.373 and porosity = 27.18%. Gradation
curve for fine aggregate conforms to zone II of IS 383:1970.
2.1.3.2 Copper slag
The physical and chemical properties of copper slag as given by the manufacturer are shown
in Table 1 and Table 2.
From Table 1, it is evident that specific gravity of copper slag is higher than the crushed
stone fine aggregates and the chemical composition of copper slag resembles that of fly ash
(Table 2).
Table 2. Chemical properties of copper slag.
Constituents % weight
Silica 26–30%
Free silica Less than 0.5%
Table 1. Physical properties of copper slag. Alumina 2%
Iron oxide 42–47%
Colour Black, glassy Calcium oxide 01–02%
Bulk density 3.25 Magnesium oxide 1.04%
PH 6.95 Copper oxides 6.1% max
Specific gravity 3.5 Sulphates 0.13%
Moisture content <0.01% Chlorides Not detected
14

2.1.3.3 Combined grading curves
Since the copper slag is used as partial replacement of fine aggregates in the study, sieve
analysis was done for different combinations of crushed stone aggregate and copper slag.
The gradation curves are shown in Fig. 1. The gradation curvesdoes not completely conform
to zone II. The gradation curve indicates that the quantity of finer materials were less in
copper slag.
2.1.4 Superplasticiser
Superplasticiser used for the production of the concrete mix was Cera Hyperplast XR–
WR40, a new generation poly-carboxylate base water reducing admixture having a specific
gravity of 1.09.
2.2 Mix proportion

Mix design for M60 grade concrete was developed according to the provisions in ACI
211.4R-93. From different trial mixes, the one shown in Table 3 was chosen for further
experimentation.
2.3 Tests on HSC

Four different mix proportions, with copper slag replacing conventional fine aggregate in
0%, 20%, 40% and 60% were considered for the study. The fresh and hardened properties
of proportioned concrete mixes were tested according to Indian standards. Workability
of the mixes were ascertained by slump test. Compressive strength, split tensile strength,
flexural strength and fracture energy of the mixes were also determined. The mixes were
designated as MC00, MC20, MC40 and MC60, where MC stands for the mix with copper
slag followed by the numeral which indicates the percentage of fine aggregate replaced
with slag.
Mode I fracture parameters were determined by conducting Three point Bend (TPB) test
on notched beams as shown in Fig. 2. Notched beam specimens having size 1000 × 150 ×
100 mm with central notch of 2 mm thick and depth of 20 mm from bottom is used. Load
deformation behaviour of the tested specimens was observed and fracture energy is calcu-
lated from the load displacement curve.
Figure 1. Sieve analysis results.
Table 3. Mix proportion for M60 concrete.
Fine Coarse
Cement aggregate aggregate Water SP
Kg/m3 Kg/m3 Kg/m3 l/m3 l/m3
479.3 642.9 1150.5 153.38 2.87
15

Figure 2. Test set up for three point beam specimen.
3.1 Workability
The slump values of M60 mix with different percentage replacement of fines with cop-
per slag is shown in Fig. 3. An increase in workability was observed with increase in
percentage replacement of conventional fines with copper slag. This may be due to the
low water absorption of copper slag. The smooth and glassy texture of copper slag may
have resulted in the reduction of friction between material particles, which in turn results
in a ball bearing effect which increases the fluidity of the mix. In addition, the surface
area of copper slag being lesser, reduced amount of paste may be sufficient for coating the
surface of copper slag, which increases the paste available for the floating of the aggregate
particles.
3.2 Compressive strength

The variation of 7 and 28 day cube compressive strength of the specimens with respect to the
percentage of slag replacement is shown in Fig. 4.
The 7-day and 28-day compressive strength showed a marginal increase with replacement
of crushed stone fine aggregate with copper slag up to a replacement level of 40% beyond
which there is reduction in strength. The 28 days average compressive strength obtained for
copper slag mix concrete with 20% and 40% copper slag exhibited 2.94% and 5.88% increase
in compressive strength when compared to control mix concrete.
The test results indicate that about 40% of natural fine aggregate can be replaced with
copper slag, without any adverse effect on compressive strength.
3.3 Split tensile strength

The values of Split tensile strengths of cylinder specimens of 0%, 20%, 40% and 60% replace-
ment of fine aggregate with copper slag and having size 150 mm × 300 mm upon testing on a
Compression testing machine is shown in Fig. 5. Split tensile strength increases by 8.9% and
30.8% for 20% and 40% replacement of fine aggregate with copper slag and for replacement.
Split tensile strength decreases by 8.6% for 60% replacement.
3.4 Flexural strength test

Flexural strength of HSC was determined and values obtained are shown in Table 4.
Flexural strength values showed 36.4% and 50% increase with respect to control specimen
for 20% and 40% replacement of copper slag respectively. Flexural strength decreases by
13.6% for 60% replacement.
The improvement in split tensile strength and flexural strength for replacement levels of 20%
and 40% can be attributed to dense packing of aggregates in these combinations, higher strength
of copper slag particles compared to crushed stone fine aggregates, less flaky particle shape
16

Figure 3. Slump test results.
Figure 4. Variation of compressive strength.
Figure 5. Split strength test results.
Table 4. Flexural strength test results.
Flexural strength
Specimen (N/mm2)
Control (0% copper slag) 5.5

20% copper slag 7.5
40% copper slag 8.25
60% copper slag 4.75
17

of copper slag aggregates. Further, the improved bond between aggregates and concrete may
have contributed to increased tensile and flexural strength. The split tensile strength and flexural
strength values indicate that fine aggregate can be replaced safely up to 40% by copper slag.
3.5 Fracture parameters

One of the most important parameter in fracture behaviour and cracking analysis is the
fracture energy representing cracking resistance and fracture toughness of concrete. For this
reason, the fracture energy is considered as a material property of concrete. Fracture energy
is the amount of energy necessary to create one unit area of crack. Energy release rate, G is
the net change in potential energy due to increment of crack extension. Fracture will occur if
energy release rate is equal to critical energy release rate (Gc).
w
G=
Aeff
where,
G = Fracture energy,
W = Total energy dissipated in the test,
Aeff = Effective area of cross section of the specimen.
Fracture toughness is an indication of amount of stress required to propagate a pre-existing
flaw. It describes the ability of a material with a crack to resist fracture. Mode I fracture param-
eters were determined by three point bending test. Load—crack mouth opening displacement
curve for various percentage of copper slag replacing fine aggregate is shown in Fig. 6 to Fig. 9.
The specimens with 20%, 40% and 60% showed 95.4%, 138% and 28% improvement in
fracture energy compared to control specimens. The fracture toughness also exhibited an
improvement of about 40% [Table 5]. This indicates that the energy required for crack ini-
tiation and propagation is more for HSC mixes with copper slag aggregates. The crack tip
Figure 6. Load Vs CMOD curve for control specimen.
Figure 7. Load Vs CMOD curve for 20% CS specimen.
18

Table 5. Mode I fracture parameters.
Mode I fracture Mode I fracture

energy (G) toughness
Specimen N/m MPa √m
Control 92.692 1.895

20% CS 181.154 2.649
40% CS 220.577 2.923
60% CS 118.846 2.145
opening load showed an improvement of 100% and the ultimate load showed an increase
of 67% for a replacement level of 40%. This improvement in fracture parameters can be
mainly attributed to the strength of aggregates and the bond between aggregates and
cement paste.
4 CONCLUSION
1. From the analysis of parameters studied, it can be seen that the properties showed consid-
erable improvement for replacements levels of 20% to 40%.
2. Workability increased as the percentage of copper slag replacing fine aggregate increased.
3. The increase in compressive strength, split tensile strength and flexural strength of
specimens with 40% replacement of fines with copper slag are 5.88%, 30.77% and 50%
respectively.
19

4. The optimum level of replacement of conventional fines by copper slag is 40%.
5. The dependability of optimum percentage of copper slag is to be studied for other grades
of concrete also.
6. The researches on copper slag replacing natural sand will give new dimension in concrete
mix design and if applied on large scale would revolutionize the construction industry by
economizing the cost and enables us to conserve natural resources.
REFERENCES
Ambily P.S., Umarani C., Ravisankar K., Prabhat Ranjan Prem, Bharathkumar B.H. and Nagesh R.
Iyer. 2015. Studies on ultra-high performance concrete incorporating copper slag as fine aggregate,
Construction and building materials, vol. 77, pp. 233–240.
Binaya Patnaik, Seshadri Sekhar, T. and Srinivasa Rao. 2014. An Experimental Investigation on
Optimum Usage of Copper Slag as Fine Aggregate in Copper Slag Admixed Concrete, International
Journal of Current Engineering and Technology, vol. 04, pp. 3646–3648.
Khalifa S. Al-Jabri, Makoto Hisada, Salem K. Al-Oraimi and Abdulla Al-Sidi H. 2009. Copper
slag as sand replacement for high performance concrete, Cement and concrete composites, vol. 31,
pp. 483–488.
Mostafa Khanzadi and Ali Benhood. 2009. Mechanical properties of high-strength concrete incorpo-
rating copper slag as coarse aggregate, Construction and building materials, vol. 23, pp. 2183–2188.
Paresh Tiwari and Anil kumar saxena. 2016. A review on partial substitution of copper slag with sand in
concrete materials, International journal of engineering research and technology, vol. 05, pp. 212–215.
Sivakumar T. and Sivaramakrishnan R. 2016. Experimental Investigation on the Mechanical Proper-
ties of Reinforced Cement Concrete using Copper Slag as Partial Replacement for Fine Aggregates,
International Journal of Current Engineering and Technology, vol. 06, pp. 834–840.
Tamil Selvi P., Lakshmi Narayani P. and Ramya G. 2014. Experimental Study on Concrete Using
Copper Slag as Replacement Material of Fine Aggregate, Civil and environmental engineering,
vol. 4, pp. 1–6.
Wei Wu, Weide Zhang and Guowei Ma. 2010. Optimum content of copper slag as fine aggregate in high
strength concrete, Materials and design, 31, pp. 2878–2883.
20

Study on effectiveness of RC beam strengthening methods

using CFRP
Preetha Prabhakaran & Glory Joseph

Division of Civil Engineering, CUSAT, Kochi, Kerala, India
ABSTRACT: One of the most prominent techniques available for repair and strengthening
of concrete structures is the use of Fiber Reinforced Polymer (FRP) system. This study
identifies the best method of installing CFRP bars in reinforced concrete beams when same
effective area of CFRP is being used at critical section of beam. The results of experimental
investigations on RC beams strengthened in flexure using various strengthening methods
such as Near Surface Mounted (NSM) and Externally Bonded Reinforcement (EBR)
methods are discussed.
Keywords: NSM, EBR, Flexural strengthening, CFRP
1 INTRODUCTION
Most of the existing structures are designed based on gravity load design (GLD); consequently,
they lack many important factors such as seismic loading and ductile detailing. In recent
years, strengthening technologies for reinforced concrete structures using FRP compos-
ites have been gaining widespread interest and growing acceptance in the civil engineering
industry. CFRP can be installed mainly in two different methods; Near Surface Mounted
(NSM) and Externally Bonded Reinforcement (EBR) on the element to enhance the flexu-
ral strength. Externally bonded FRP (EBR) can be acknowledged as a method of binding
a sheet or a plate with an epoxy on the outer surface of a structure. Near surface mounted
(NSM) bars has emerged as a new strengthening technique in which the external reinforce-
ment is embedded into the section in grooves with an epoxy adhesive. [4][1][2]
Even though NSM method has been identified as an effective method of installing CFRP
in qualitative terms, experimental values have not been provided yet to verify the effectiveness
of NSM method in terms of flexural strengthening when same effective area of CFRP is used
as in EBR technique.
2 FLEXURAL STRENGTHENING
2.1 NSM method vs. EBR method

NSM method is been adopted where FRP or steel reinforcements are inserted into grooves
with an epoxy adhesive cut into the structural members. The EBR technique involves the
external bonding of strengthening materials such as steel plates or fiber reinforced poly-
mer (FRP) laminates. However, this technique suffers from the high possibility of premature
failure such as debonding of longitudinal laminates, delamination and other types of pre-
mature failure [2] [10]. Researches carried out on reinforced concrete (RC) flexural mem-
bers strengthened with EBR showed low efficiency of this technique, due to premature FRP
debonding failure. The FRP materials used in EBR technique can be subjected to physical
and fire damage [8]. In order to overcome these drawbacks, and to ensure higher durability,
21

NSM technique was recently introduced as an alternative strengthening technique to the
EBR technique for increasing flexural strength of reinforced concrete members. [8] [6]
3 EXPERIMENTAL PROGRAMME
In this study, the test program consists of two series of tests which includes casting,
instrumentation, and testing nine RC beams having rectangular cross-section. The beams
were designed as under reinforced beams to initiate failure in flexure. The 28 days average
compressive strength of the concrete was 36.2 MPa. The RC beams have length of 1000 mm
with cross section of 150 mm width, 200 mm depth and effective span of 700 mm. Each series
was composed of a Control beam (CB), and three beams for each investigated strengthening
technique. In this paper an attempt is made to evaluate the flexural performance of RC beams
using NSM and EBR techniques with same effective area of CFRP at mid span section. The
properties of CFRP bar, CFRP sheets and epoxy are presented in Table 1.
3.1 Flexural strengthening setup

The dimensions and details of the tested control beam is given in Fig. 1. All the flexural
strengthening beams have two longitudinal steel bars (Fe500 grade) of 10 mm diameter (2Ø10)
at the bottom, and 2Ø10 at the top with a clear cover of 25 mm at the top and 30 mm at the
bottom of the beam (see Fig. 1). The transverse reinforcement is composed of steel stirrups of
8 mm diameter (Ø8) with a constant spacing of 100 mm in order to avoid shear failure.
For NSM method, NSM bars are of diameters 10 mm and 12 mm in order to study the
influence C/s area of NSM bar. The designation CB stands for control beam, and NSM
Table 1. Material specifications.
Material Parameters Value
CFRP bar Tensile strength 2172 MPa

Modulus of Elasticity 124 GPa
% elongation 1.75%
Density 0.1116 kg/m
CFRP sheets Fiber thickness 0.25 ± 0.03 mm
Sheet width 1000 mm
Tensile strength 3450 MPa
Modulus of Elasticity 230000 MPa
% elongation 1.5%
Density 1.8 g/cc
Epoxy (Polyester resin) Compressive Strength 87 MPa
Tensile Strength 11 MPa
Flexural Strength 19 MPa
Compressive modulus 13 GPa
Density 1880 kg/m3
Araldite epoxy resin Specific Gravity 1.17
Viscosity at 25ºC (Pa.s) 50
Colour Creamy viscous
Figure 1. Detailing of RC beam specimen ( CB—Flexure).
22

Figure 2. NSM strengthened beam (BD10G20 & BD12G25).
Figure 3. U wrapped RC beam (UBD12G20).
Figure 4. EBR strengthened beam (EB01).
Table 2. Specifications of tested beams.
C/s area
Sl. Method of of CFRP
No. Beam ID strengthening Description (mm2)
1 CB Nil Control Beam –

2 BD10G20 NSM Strengthened beam with 10 mm dia CFRP 78.5
bar and groove size 20 mm
3 BD12G20 Strengthened beam with 12 mm dia CFRP 113
4 BD12G25 Strengthened beam with 12 mm dia CFRP 113
5 UBD12G20 U wrapping CFRP U wrapped beam with 12 mm dia CFRP 263
bar, 20 mm groove size and 1.5 mm thick,
100 mm wide CFRP sheet
6 UBD12G25 CFRP U wrapped beam with 12 mm dia CFRP 263
bar, 25 mm groove size and 1.5 mm thick,
100 mm wide CFRP sheet
8 EB01 EBR Strengthened beam with 1 layer of CFRP 78
sheet of 1.5 mm thick and 700 mm length
9 EB02 Strengthened beam with 2 layer of CFRP 135
sheet of 3 mm thick and 600 mm length
23

strengthened beam was designated as BD10G20, where B stands for NSM beam, D stands
for the diameter i.e. 10 mm, G stands for groove size i.e. 20 × 20 mm. Fig. 2 includes the cross
section of the beams for NSM flexural strengthening. In order to study the effect of groove
width, depth and the net distance between a groove and the beam edge which can influence
the bond performance and hence the structural behaviour of NSM strengthened beam, two
sizes of grooves are considered such as 20 mm × 20 mm and 25 mm × 25 mm respectively.
In order to avoid debonding failure, two RC specimen samples are provided with CFRP
U – Wrap at the ends of the groove for NSM beams. The influence of CFRP U – wrap as
end anchorage is studied by providing CFRP sheet of thickness 1.5 mm and width 100 mm
as U – wrapping at the ends of the NSM groove as shown in Fig. 3. For bonding CFRP sheet
onto the RC beam specimen, araldite epoxy resin has been used. NSM strengthened beam
with CFRP U – Wrapping is designated as UBD12G20, UB stands for U – wrapped NSM
strengthened beams.
For EBR method, CFRP sheets of thickness 1.5 mm & 3 mm and width of 52 mm &
45 mm respectively are externally bonded to the bottom surface of the beam using Araldite
epoxy resin. EBR strengthened beam was designated as EB01 and EB02, where EB stands
for EBR beam, 01 and 02 stands for one layer of CFRP or two layers of CFRP sheets respec-
tively. Fig. 4 shows the cross section of the beams for EBR flexural strengthening.
A summary of test beam specifications is presented in Table 2.
4.1 Failure modes

Reinforced concrete elements strengthened with FRP systems are observed with additional
failure modes compared with reinforced concrete elements that are not strengthened with
FRP systems. Most of the beams are observed with flexural end debonding. Fig. 5 shows
the various failure patterns obtained for the tested beams. The control beam CB failed at a
load of 82 kN by concrete crushing after yielding of steel reinforcement. The formation of
the cracks followed a typical crack pattern of flexural members as shown in Fig. 5. The first
visible crack was observed at the mid span at a total load of about 16.5 kN. By increasing the
loads, new flexural cracks started to form and distributed along the length of the beam mark-
ing extensive yielding of steel reinforcement followed by crushing of concrete in compression.
All strengthened beams showed similar behaviour as that observed for the control beam up to
first crack load. For beams flexurally-strengthened with NSM reinforcement, the common fail-
ure mode observed is concrete crushing. The EBR strengthened beams failed because of strip
end debonding due to loss in the composite action between the bonded CFRP and RC member.
This may occur since carbon fibers have very high strength in tension relative to the concrete
and to the adhesive that binds them to the concrete. In U wrapped specimens (UBD12G20
& UBD12G25), flexural failure combined with shear failure also have been observed. Failure
mode such as concrete cover separation, crushing of compressive concrete, plate end interfacial
debonding and shear failure have been reported by other researchers [1][11].
Test results and failure modes of various beams are given in Table 3.
4.2 Influence of cross sectional area of NSM bar

Failure of Beam BD10G20, strengthened with 10 mm dia CFRP rod with 20 mm groove
size occurred under an applied load of 98 kN, corresponding to a 19.5% increase in capacity
with respect to CB. Beam BD12G20, strengthened with 12 mm dia CFRP rod with 20 mm
groove size failed at load of 110 kN, which indicated a 34.15% increase over CB. Both beams
failed by flexural failure and crushing of concrete. Longitudinal splitting cracks developed
in the epoxy cover and lead to the loss of bond between the NSM reinforcement and epoxy.
Also, as the diameter of bar increases, the bond between polyester resin and the reinforcing
bar decreases. Thus it reduces the bond stress developed and hence in case of bar diameter of
24

Figure 5. Failure pattern of (a) CB (b) BD10G20 (c) BD12G20 (d) UBD12G20 (e) BD12G25
(f ) UBD12G25.
Table 3. Load deflection characteristics.
Pcr Pu δu ∆Pu
Beam ID kN kN mm % Mode of failure
CB 16.5 82 30 – Flexural failure & concrete crushing

BD10G20 26 98 32 19.5
BD12G20 28 110 34.2 34.2 Flexural failure & debonding
BD12G25 27.6 108 33 31.70
UBD12G20 25.3 115 24.86 40.2 Flexural & shear failure
UBD12G25 17.6 112 24.02 36.6
EB01 21 96.9 26.62 18.2 Flexural failure & debonding
EB02 22.5 97.5 24.96 18.9
Pcr – first crack load; Pu – Ultimate Load; δu – Deflection at ultimate load; ∆Pu – % ultimate load increase
over control beam.
Figure 6. Influence of C/S area of NSM bar.
12 mm, cracking of polyester resin did not occur. Due to this, the increase in load carrying
capacity of the beam BD12G20 is 14.7% when compared with BD10G20.
From Fig. 6, it is clear that at higher loads, the deflection of the strengthened beam is decreased
compared to control beam though final deflection at ultimate load is higher in strengthened
beam. This is in agreement with lesser number of cracks and finer cracks as shown in failure
pattern in Fig. 5. At ultimate load, the deflection of the beam increased by 6.7% and 14% in case
of BD10G20 and BD12G20 respectively when compared with the control beam.
25

4.3 Influence of CFRP U wrapping in NSM
CFRP U – Wrapping was provided as end anchorage in strengthened beams. In case of NSM
strengthened beam with 12 mm diameter bar (BD12G20) debonding failure occurred at polyes-
ter concrete interface. Hence CFRP U – wrapping was provided for a width of 100 mm nearer
to the end of NSM strengthened region to study its influence in strengthening of RC beam and
prevention in debonding failure. It was observed that UBD12G20 and UBD12G25 failed under
shear, with a combination of major shear cracks and minor fine flexural cracks as shown in
Fig. 5. This implies that providing end anchorage in the form of U – Wrapping in combination
with NSM technique can significantly enhance the flexural capacity of the beam. Test results of
the U – Wrapped specimen showed an increase in load carrying capacity by 40.2%, and 36.6% in
case of UBD12G20 and UBD12G25 respectively when compared with control beam specimen.
Whereas specimen without U – wrapping showed only 32% to 34% increase in strength. How-
ever Fig. 7 shows brittle failure in U – wrapped specimen compared to control beam.
4.4 Influence of number of layers of CFRP in EBR

The ultimate load of the specimen bonded with CFRP sheets (EB01 & EB02) was increased
by 18.17% and 18.9% respectively. This indicates that with the addition of one more layer
of CFRP on the tension side of the beam, the load carrying capacity has not been increased
because of the debonding failure of layers of CFRP.
Fig. 8 shows that, as the number of CFRP layers is increased, the deflection of the beam
has been reduced by 22% and 30.2% in case of EB 01 and EB 02 respectively when compared
with the control beam. But the problem of debonding is a major concern when CFRP layers
are attached to the bottom side of the beam.
4.5 Comparison between NSM and EBR

The effectiveness in terms of flexural strengthening of near surface mounting (NSM) method
is compared with externally bound reinforcement (EBR) method with the usage of same cross
sectional area of CFRP at midspan. The beams BD10G20 & EB01 and BD12G20 & EB02
are strengthened with CFRP with cross sectional area of 78 mm2 and 124 mm2 respectively.
Strengthening with EBR requires less man hours to install while the NSM method offers more
Figure 7. Influence of CFRP U wrapping.
Figure 8. Influence of number of layers of CFRP.
26

protection against loss of strengthening by integrating the additional reinforcement within the
member, which proved effective in preventing splitting or separation. In EBR strengthened
beams the main failure mode was debonding which reduces the efficiency of strengthening.
From Table 3, the results obtained exhibits 15.3% increment of load carrying capacity for
BD12G20 in NSM method Vis a Vis EB02 in EBR method when compared with CB. Though
the increase is very marginal in NSM beam with low percentage of CFRP, the ultimate load
carrying capacity of NSM beam with higher CFRP percentage is more than that of EBR
strengthened beam with the same effective cross-sectional area of CFRP. Also it is clear
that at higher loads, the deflection of the EBR strengthened beam is decreased compared to
control beam though final deflection at ultimate load is higher in NSM strengthened beam.
Thus the present study has shown that using NSM bars greatly increases the flexural capacity
over an externally bonded FRP system with the usage of same cross sectional area of CFRP.
5 CONCLUSIONS
The major conclusions derived from the experimental study are as follows:
The influence of crossectional area of NSM CFRP in flexural strengthening is studied
and it is found that the beams strengthened with 10 mm and 12 mm diameter bars showed
an increase in load carrying capacity by 19.51% and 34.15% respectively compared to
unstrengthened control beam.
Beams strengthened with CFRP NSM reinforcement along with U—wrapping as end
anchorage can increase the efficiency of the strengthening technique to an extent of 6%
compared to specimens without U wrapping.
The ultimate load of the EBR specimen bonded with CFRP sheets is increased by 18%
compared to control beam. With the addition of one more layer of CFRP on the tension
side of the beam, the load carrying capacity has been increased only marginally because of
the debonding failure observed.
The flexural capacity of NSM strengthened beam with CFRP cross-section is increased
as much as 15% compared to EBR strengthened beam with similar effective crossectional
area of CFRP.
REFERENCES
Ahmed Ehsan et al. 2011. Flexural performance of CFRP strengthened RC beams with different degrees
of strengthening schemes. International Journal of the Physical Sciences 6(9): 2229–2238.
Balamuralikrishnan R. & Antony C. 2009. Flexural Behavior of RC Beams Strengthened with Carbon
Fiber Reinforced Polymer (CFRP) Fabrics. The Open Civil Engineering Journal 3: 102–109.
Barros J.A.O. and Dias S.J.E. 2006. Near surface mounted CFRP laminates for shear strengthening of
concrete beams. Cement & Concrete Composites for Construction Elsevier 28: 276–292.
Bilotta A. et al. 2011. Bond Efficiency of EBR and NSM FRP Systems for Strengthening Concrete
Members. Journal of Composites for Construction 15: 757–772.
Eliane K. et al. 2007. Flexural Strengthening of RC-T Beams with Near Surface Mounted (NSM) FRP
reinforcements. University of Patras, Patras, Greece 16–18.
George Varughese Abilash & Manikandan T. 2015. Flexural Retrofitting of RC Beams using Extra
Rebars and U-Wraps. International Journal of Science and Research 5: 2305–2309.
Khalifa M. Ahmed. 2016. Flexural performance of RC beams strengthened with near surface mounted
CFRP strips. Elsevier 55: 1497–1505.
Laraba Abdelkrim et al. 2014. Structural Performance of RC Beams Strengthened with NSM-CFRP.
Proceedings of the World Congress on Engineering 2.
Moshiur Rahman, Md. et al. 2015. Effect of adhesive replacement with cement mortar on NSM
strengthened RC Beam. University of Malaya 61–72.
Shukri A.A. et al. 2016. Behaviour of precracked RC beams strengthened using the side-NSM tech-
nique. Elsevier 123: 617–626.
Singh S.B. et al. 2014. Experimental and parametric investigation of response of NSM CFRP-
strengthened RC Beams. ASCE J. Compos. Constr. 18: 1–11.
27

Performance of nano-SiO2 admixed concrete under saline

environment
M. Daniyal & S. Akhtar

Department of Civil Engineering, Aligarh Muslim University, Aligarh, India
A. Azam
Department of Applied Physics, Aligarh Muslim University, Aligarh, India
ABSTRACT: The effect of nano-SiO2 on the mechanical strengths of concrete and

corrosion rate of embedded steel under different mixing and curing conditions was inves-
tigated. Twelve types of concrete mixes were made by altering the content of nano-SiO2
(1%, 3% and 5% by weight of cement) and type of mixing and curing water. Potable and
saline (3.5% NaCl) water were used for mixing and curing for different concrete mixes. The
mechanical strengths were found to increase for specimens mixed and cured in saline water at
the early ages, while, a reduction in the respective strengths was observed for later ages. How-
ever, it was observed that the addition of nano-SiO2 increases the strength of concrete and
reduces the corrosion rate of embedded steel as compared to reference mixes. This is because
the nano-SiO2 fills the tiny voids between the hydration products and thus, reduces porosity,
which leads to a compact structure of concrete.
1 INTRODUCTION
Steel reinforced concrete is considered as an ideal composite material and thus, it has been
used extensively for the construction of all types of structures. This is mainly because of the
high compressive strength and excellent tensile strength contributed by concrete and steel
respectively. However, its structural and durability performance is affected when subjected
to aggressive environments (Shetty, 2013). The corrosion of embedded steel with subsequent
spalling and delamination of concrete cover is one of the most common causes of deterio-
ration, particularly when using admixtures containing chloride, unwashed sea sand or the
structure is exposed to seawater or de-icing salt. Hence, the choice of materials for concrete
mix, type of steel and appropriate reinforcement detailing are crucial parameters in con-
structing durable steel reinforced concrete structures in the marine environment (Neville &
Brooks, 1994).
The main chemical ingredients of seawater are chloride, sodium, magnesium, calcium and
potassium ions. Seawater contains dissolved salts of about 35,000 ppm and sodium chloride
is the main component of dissolved salts (McCoy, 1996). Seawater is alkaline in nature with a
pH value ranging between 7.5 and 8.4. The corrosion process of embedded steel in concrete
arises when the pH value becomes lower than 11. Hence, it is necessary to supply alkalinity
into steel reinforced concrete structures, particularly when the structure is exposed to an
extremely severe environment (Gani, 1997). Chloride ions, present in seawater, can enter the
concrete and accelerate the corrosion process of steel reinforcements. This is the most damag-
ing effect of seawater on steel reinforced concrete structures (Aburawi & Swamy, 2008). The
sodium and potassium ions present in seawater may increase the alkali aggregate reaction.
Magnesium and sulfate ions may deteriorate the cement paste (Uddin et al., 2004).
Various investigations have revealed the effects of saline water on the mechanical strengths
of cementitious composites and it was observed that saline water is not fit for concrete
29
CH005_160.indd 29 7/9/2018 5:21:19 PM

construction works (Akinkurolere et al., 2007). However, a study reported that concrete with
loam as an admixture can enhance the mechanical strength durability properties of concrete
under normal as well as saline exposure conditions (Naghoj & Abdel-Rahmna, 2005). In
another study, it was observed that the compressive strength of concrete is improved with
the replacement of some amount of fine aggregate by ground blast furnace slag and ground
basaltic pumice under seawater attack (Binici et al., 2008).
Recently, nano-SiO2 has gained significant attention in the construction industry primarily
because of its pozzolanic property and large surface area to volume ratio. The addition of
nano-SiO2 was found to influence the fresh as well as hardened properties of cementitious
composites significantly, that primarily depend on nano-SiO2content, water-binder ratio, size
of nano-SiO2, type of dispersion technique and curing time. Regarding fresh properties, it has
been observed that the inclusion of nano-SiO2 considerably reduced the Initial Setting Time
(IST) as well as the Final Setting Time (FST) of cementitious composite. This may be due to
its (nano-SiO2) chemical reactivity (pozzolanic activity) or higher specific surface area, which
accelerate cement hydration and ultimately reduce the setting time of composite (Qing et al.,
2007; Senff et al., 2009; Zhang & Islam, 2012; Zhang et al., 2012; Ltifi et al.,2011). As far as
mechanical and durability properties of nano-SiO2 incorporated cementitious composites are
concerned, the compressive strength of 5% nano-SiO2 incorporated cement paste was found to
be increased by 8%, 41%, 25% and 15% at the ages of one day, three days, 28 days and 60 days
respectively; and the bond strength was found to be increased by 88% at the age of 28 days
as compared to plain cement paste (Qing et al., 2007). The three day compressive strength of
nano-SiO2 incorporated high volume fly ash concrete was found to be increased by 81% as
compared to high volume fly ash concrete and also, a similar trend was reported at later stages
(Li, 2004). Also, it has been found that the cement could be replaced by nano-SiO2 (average
particle sizes of 15 and 80 nm) up to 2.0% with improved performance. However, the optimal
level of replacement was found to be 1.0% and 1.5% for 15 nm and 80 nm size respectively
(Givi et al., 2010). On replacement of cement with 1% of nano-SiO2, both the compressive
and flexural strength of concrete was found to be increased by 12.31% and 4.21%, respectively
and the porosity and chloride permeability was found to be decreased by 6.93% and 18.04%,
respectively. However, a negative effect was observed on an increase of replacement level by
3% (Zhang & Li, 2011).Similar trends of results were obtained in various investigations (Jo
et al., 2007; Gaitero et al., 2008; Zapata et al., 2013; Kawashima et al., 2013).Therefore, in
order to usenano-SiO2 in the construction sector effectively, more research and investigations
are required to determine various facts like performance of nano-SiO2 and its reactive mecha-
nism with composite materials. In this study, the effect of nano-SiO2 content, mixing and cur-
ing water on the properties of concrete, specifically, compressive strength, flexural strength,
tensile strength and corrosion behavior of embedded steel are addressed. Therefore, an experi-
mental investigation was conducted using various mixes of concrete with different amounts of
nano-SiO2 and potable water and saline water for mixing and curing.
2 EXPERIMENTAL INVESTIGATION
2.1 Concrete mix and materials

In this study, 12 types of concrete mixes were prepared. First, second, third and fourth type
of mixes were prepared using potable water with 0%, 1%, 3% and 5% of nano-SiO2 content
by weight of cement, respectively. Also, the samples were cured under potable water, thus des-
ignated as Potable-Potable (PP) mixes. Further, fifth, sixth, seventh and eighth type of mixes
were prepared using potable water with 0%, 1%, 3% and 5% of nano-SiO2 content by weight
of cement, respectively, but the samples were cured under saline water, therefore designated
as Potable-Saline (PS) mixes. Furthermore, ninth, tenth, eleventh and twelfth type of mixes
were prepared using saline water with 0%, 1%, 3% and 5% of nano-SiO2 content by weight
of cement, respectively, and the samples were cured with saline water, hence designated as
Saline-Saline (SS) mixes. The concrete specimens were tested at the ages of 7, 28 and 90 days
for compressive flexural and split tensile strengths. Moreover, electrochemical tests were
30

Table 1. Concrete mix proportions.
Cement FA CA NS Mixing Curing

Mix kg/m3 kg/m3 kg/m3 kg/m3 water water
PP 300 600 1200 0 Potable Potable

S1PP 300 600 1200 3 Potable Potable
PS 300 600 1200 0 Potable Saline
S1PS 300 600 1200 3 Potable Saline
S5SS 300 600 1200 15 Saline Saline
SS 300 600 1200 0 Saline Saline
conducted at the age of 90 days in order to determine the corrosion rate of embedded steel
in concrete for all the mixes. All the specimens were cast using Ordinary Portland Cement
(OPC). The mix ratio was taken as 1:2:4. Coarse aggregates with maximum nominal size
of 12.5 mm were used for all mixes. Locally available river sand was used as fine aggregate.
The nano-SiO2,with an average particle size of 15 nm, was used as an admixture. The water/
cement ratio was taken as 0.5. The cement content used was 300 kg/m3. Potable water with
a pH value of about 7.12 and saline water (sodium chloride 3.5% by weight of water) with
a pH value of about 7.31 were used for mixing and curing. All the concrete specimens were
cast and demolded after 24 hours and were kept under specified curing water up to the testing
days. Moreover, three specimens for each condition were prepared in order to understand the
variability of the test results. Quantities of the materials used for the all concrete mixes such
as Fine Aggregate (FA), Coarse Aggregate (CA), cement and so on, are shown in Table 1.
2.2 Specimen specifications and testing methods

Standard concrete cubes of size 150 × 150 × 150 mm were used for determining the com-
pressive strengths and the tests were carried out as per IS: 516-1959. Concrete beams of
size 500 × 100 × 100 mm were used for measuring the flexural strengths and the tests were
conducted as per IS:516-1959. Cylindrical concrete specimens of size 150 mm diameter and
300 mm height were used for determining the split tensile strengths and the tests were carried
out as per IS-5816-1999.
For determining the corrosion rate of embedded steel, cylindrical concrete specimens of
size 50 mm diameter and 100 mm height, at the center of which a single mild steel bar of
diameter 10 mm and length 95 mm was placed, were used. The bar was positioned in such a
way that it projected from the top of the mold by 15 mm and therefore provided a uniform
concrete cover of 20 mm from the sides and bottom. The projected part of the steel bar
was used for electrical connection and was wrapped until test time. These specimens were
monitored continuously with biweekly wetting and drying cycles in order to accelerate the
corrosion process. The cyclic sweep measurements were performed using the Gill AC poten-
tiost at supplied by ACM (Applied Corrosion Monitoring) instruments. The most common
three-electrode electrochemical cell was used in which the specimens served as the Work-
ing Electrodes (WE), a Saturated Calomel Electrode (SCE) and platinum electrode served
as the Reference Electrode (RE) and the Counter Electrode (CE), respectively. A Working
Sense (WS) was also connected to a working electrode. Potable water or saline water (con-
taining 3.5% NaCl) was used as an electrolyte of the cell depending on the curing condi-
tions. In this electrochemical cell, the WE and CE carry the current; and the WS and RE
measure potential (voltage).The cyclic sweep measurements were carried out at the exposure
duration of 90 days with start potential −300 mV, reverse potential 300 mV and scan rate of
31

60 mV/minute. Then, the corrosion parameters such as corrosion potential (Ecorr), corrosion
current density (Icorr) and corrosion rate (CR) were determined by means of Tafel extrapola-
tion of the polarization curves.
3.1 Mechanical strengths

It was observed that the concrete specimens mixed and exposed in saline water had darkened
surfaces with reference to the specimens mixed and exposed in potable water. Also, salt
deposits were seen on the surfaces of concrete specimens exposed under saline water. The
average value of investigational outcomes for mechanical strengths of concrete regarding
compressive, flexural and split tensile strengths are shown in Figures 1 to 9.
3.1.1 Effect of mixing and curing water

Significant enhancements in early mechanical strengths were observed for SS and PS mixes,
with respect to the PP mixes, at the age of seven days. However, at the ages of 28 days and
Figure 1. Effect of nano-SiO2 on compressive Figure 2. Effect of nano-SiO2 on compressive

strength of PP mixes. strength of PS mixes.
Figure 3. Effect of nano-SiO2 on compressive Figure 4. Effect of nano-SiO2 on flexural
strength of SS mixes. strength of PP mixes.
Figure 5. Effect of nano-SiO2 on flexural Figure 6. Effect of nano-SiO2 on flexural
strength of PS mixes. strength of SS mixes.
32

Figure 7. Effect of nano-SiO2 on split tensile Figure 8. Effect of nano-SiO2 on split tensile
strength of PP mixes. strength of PS mixes.
Figure 9. Effect of nano-SiO2 on split tensile strength of SS mixes.
90 days, higher mechanical strengths were observed for the PP mixes as compared to the SS
and PS mixes. Thus, the rate of strength gain of SS and PS mixes was faster than PP mixes at
the early ages. This showed that the saline water accelerates the hydration of cement. How-
ever, a reduction in strengths of SS and PS mixes at the later age were reported, which may
be due to the formation of salt crystallization in concrete and thus, affecting the strength.
3.1.2 Effect of nano-SiO2

The effect of nano-SiO2 on the compressive strength of concrete for PP, PS and SS mixes are
shown in Figures 1, 2 and 3, respectively. As the figures specify, the compressive strength of
all the specimens enhanced with an increase in the ages irrespective of their mixing and cur-
ing conditions and reached peak values at 90 days. Also, an upward trend in strength with
increasing the quantity of nano-SiO2 was observed. Similar trends were observed for flexural
and split tensile strengths as shown in Figures 4 to 9. The results demonstrate that addition
of nano-SiO2 to the concrete mixes enhanced the mechanical strengths at all stages of curing
and this enhancement was more evident at the later stages of hydration. There may be three
reasons for such enhancements: a) nano-SiO2 can be uniformly distributed in the mix and
will serve as a nucleus to bond with hydrates of cement; b) nano-SiO2 fills the voids between
cement particles, reducing the porosity; c) nano-SiO2 may accelerate the hydration process
owing to its high pozzolanic property.
3.2 Corrosion rate of embedded steel

The polarization curves (plot of potential versus current density) of all the concrete mixes
containing the steel bar are shown in Figures 10 to 12, according to the mixing and curing
conditions. Corrosion kinetic parameters, as listed in Table 2, were obtained from polariza-
tion curves with the use of the Tafel extrapolation technique. It was observed that the CR
of PP mixes was significantly low compared to the PS and SS mixes, and the SS mixes were
found to be the most vulnerable. This is due to the presence of chloride ions which accumu-
late on the surface of steel and destroy the passive protective film of steel, thereby accelerat-
ing the corrosion process. However, the nano-SiO2 admixed cementitious composites showed
a lower CR compared to the reference mixes. Also, it was observed that the CR was found
to decrease with an increasing content of nano-SiO2. This is mainly because the nano-SiO2
33

Figure 10. Polarization curves for PP mixes. Figure 11. Polarization curves for PS mixes.
Figure 12. Polarization curves for SS mixes.
Table 2. Corrosion parameters.
Ecorr Icorr CR
Specimen mV mA/cm2 × 10-5 mm/year × 10-3
PP -413.5 464.2 53.8

S1PP -328.2 159.4 18.5
S3PP -186.6 80.7 9.4
S5PP -181.2 33.9 3.9
PS -543.3 820.4 95.1
S1PS -473.3 288.6 33.5
S3PS -329.5 123.2 14.3
S5PS -247.9 111.3 12.9
SS -701.4 12779.2 1481.1
S1SS -673.6 7902.6 916.9
S3SS -632.8 5352.9 620.4
S5SS -598.8 4483.7 519.6
fills the tiny voids between the hydration products and improves the bonding capacity, lead-
ing to a compact structure with improved density and reduced porosity. Thus, the ingress of
chloride ions is significantly reduced.
4 CONCLUSIONS
The following conclusions may be drawn on the basis of the present investigations:
1. Concrete mixes prepared as well as cured in saline water have higher mechanical strengths
than concrete mixes prepared and cured in potable water at the early ages. However, the
later age strengths of concrete mixes mixed and cured in potable water were found to be
significantly higher than the concrete mixes mixed and cured in saline water. This reduc-
tion in strength is possibly due to the formation of salt crystals in concrete.
34

2. The nano-SiO2 admixed concrete mixes showed higher mechanical strengths compared
to the reference concrete mixes. Also, the strength increases with an increasing content of
nano-SiO2. This enhancement in strength is possibly due to the fact that the nano-SiO2
fills the voids between cement hydrates, thus, reducing the porosity which leads to a com-
pact structure.
3. The CR of specimens prepared using saline water and cured in saline water was found to
be much higher than specimens prepared and cured in potable water. This may be because
the steel is directly attacked by chloride ions. However, the CR of specimens prepared
using potable water and cured in saline water was found to be lower than the specimens
prepared and cured in saline water. This is due to the fact that, in this case, the steel is not
directly exposed to chloride ions and thus, the CR depends on the ingress of chloride ions.
4. The nano-SiO2 admixed specimens showed a lower CR compared to the reference speci-
mens under different mixing and curing conditions. Furthermore, the CR decreases con-
siderably with an increasing amount of nano-SiO2. This reduction in CR is possibly due
to the fact that the nano-SiO2 decreased the porosity of composites and thus, reduced the
ingress of chloride ions.
5. Further study and work is needed on this topic of using saline water for durable cemen-
titious composites in future construction work, as fresh water sources are decreasing
globally.
ACKNOWLEDGMENTS
M. Daniyal, is thankful to the Council of Scientific and Industrial Research (CSIR), Human
Resource Development Group, Government of India, for providing financial assistance in
the form of a Senior Research Fellowship (SRF).
REFERENCES
Aburawi, M. & Swamy, R.N. (2008). Influence of salt weathering on the properties of concrete. The
Arabia Journal for Science and Engineering, 33 (N 1B), 105–115.
Akinkurolere, O.O., Jiang, C. & Shobola, O.M. (2007). The influence of salt water on the compressive
strength of concrete. Journal of Engineering Applied Science, 2(2), 412–415.
Binici, H., Aksogan, O., Gorur, E.B., Kaplan, H. & Bodur, M.N. (2008). Performance of ground blast
furnace and ground basaltic pumice concrete against seawater attack. Construction and Building
Materials, 22(7), 1515–1526.
Gaitero, J.J., Campillo, I. & Guerrero, A. (2008). Reduction of the calcium leaching rate of cement paste
by addition of silica nanoparticles. Cement and Concrete Research, 38, 1112–1118.
Gani, M.S.J. (1997). Cement and concrete (First)., London, UK: Chapman & Hall.
Givi, A.N., Rashid, S.A., Aziz, F.N.A. & Salleh, M.A.M. (2010). Experimental investigation of the size
effects of SiO2 nano-particles on the mechanical properties of binary blended concrete. Composites:
B, 41, 673–677.
IS: 516-1959, Indian Standard for methods of tests for strength of concrete, Reaffirmed 1999.
IS:5816-1999, Indian Standard for splitting tensile strength of concrete-method of test, Reaffirmed 2004.
Jo, B.W., Kim, C.H., Tae, G.H. & Park, J.B. (2007). Characteristics of cement mortar with nano-SiO2
particles.Construction and Building Materials, 21, 1351–1355.
Kawashima, S., Hou, P., Corr, D.J. & Shah, S.P. (2013). Modification of cement-based materials with
nanoparticles. Cement and Concrete Composites, 36, 8–15.
Li, G. (2004). Properties of high-volume fly ash concrete incorporating nano-SiO2. Cement and Concrete
Research, 34, 1043–1049.
Ltifi, M., Guefrech, A., Mounanga, P. & Khelidj, A. (2011). Experimental study of the effect of addition
of nanosilica on the behaviour of cement mortars. Procedia Engineering, 10, 900–905.
McCoy, W.J. (1996). Mixing and curing water for concrete. Significance of tests and properties of
concrete and concrete making materials, STP 169-A. Philadelphia, PA: American Society for Testing
and Materials, 515–521.
35

Naghoj N.M. & Abdel-Rahmna, N. (2005). Enhancing the performance of concrete subjected to salty
seawater. In Admixtures–Enhancing Concrete Performance, The International Conference (pp. 35–40),
Dundee, Scotland, UK. London: Thomas Telford.
Neville, A.M. & Brooks, J.J. (1994). Concrete technology. England: Longman Scientific and Technical.
Qing, Y., Zenan, Z., Deyu, K. & Rongshen, C. (2007). Influence of nano-SiO2 addition on properties
of hardened cement paste as compared with silica fume. Construction and Building Materials, 21,
539–545.
Senff, L., Labrincha, J.A., Ferreira, V.M., Hotza, D. & Repette, W.L. (2009). Effect of nano-silica on
rheology and fresh properties of cement pastes and mortars. Cement and Concrete Research, 23,
2487–2491.
Shetty, M.S. (2013). Concrete technology theory and practice. India: S. Chand and Company Pvt. Ltd.
Uddin, T., Hidenori, H. & Yamaji, T. (2004). Performance of seawater-mixed concrete in the tidal
environment. Cement and Concrete Research, 34(4), 593–601.
Zapata, L.E., Portela, G., Suarez, O.M. & Carrasquillo, O. (2013). Rheological performance and
compressive strength of superplasticized cementitious mixtures with micro/nano-SiO2 additions.
Construction and Building Materials, 41, 708–716.
Zhang, M. & Islam, J. (2012). Use of nano-silica to reduce setting time and increase early strength of
concretes with high volumes of fly ash or slag. Construction and Building Materials, 29, 573–580.
Zhang, M., Islam, J. & Peethamparan, S. (2012). Use of nanosilica to increase early strength and reduce
setting time of concretes with high volumes of slag. Cement and Concrete Composites, 34, 650–662.
Zhang, M. & Li, H. (2011). Pore structure and chloride permeability of concrete containing nano-
particles for pavement. Construction and Building Materials, 25, 608–616.
36

Strengthening of structural elements using Fiber-Reinforced

Polymer (FRP)—a review
B. Vijayasree & Mini Soman

ABSTRACT: In India most structures are designed for gravity loading as per IS 456:2000.
These structures are susceptible for damage during earthquakes. During a severe earthquake,
the structure is likely to undergo inelastic deformation and has to depend on ductility and
energy absorption capacity to avoid collapse. Such buildings designed for gravity loading
need to be strengthened to increase strength, stiffness and ductility. Some recently developed
Fiber-Reinforced Polymer (FRP) techniques can play a vital role in structural repairs, seismic
strengthening and retrofitting of existing buildings, whether damaged or undamaged.
Jacketing is the most popularly used method for the strengthening of building structural
elements. This paper presents a review of FRP jacketing on different structural elements of
a building.
1 INTRODUCTION
Strengthening of existing reinforced concrete structures is now a major construction activity

all over the world. The Reinforced Concrete (RC) structures constructed across the world
are often found to exhibit distress and suffer damage, even before service life is over, due
to several causes such as earthquakes, corrosion, overloading, change of codal provisions,
improper design, faulty construction, explosions and fire. With the advent of revised design
and detailing codes, and an increase in the earthquake vulnerability level of many regions, the
existing structures need retrofitting and strengthening. Various types of construction indus-
try strengthening materials, such as steel plates, ferrocement, and Fiber-Reinforced Polymer
[FRP] are available to be used for the jacketing of affected components, the most common
being steel jackets. These types of jackets increase the weight and dimensions of the structural
elements. FRP based strengthening has become attractive when compared to others, due to
its light weight, high strength and stiffness, corrosion resistance and easier implementation.
2 FIBER REINFORCED POLYMER (FRP)
Fiber-reinforced polymer (also called fiber-reinforced plastic) is a composite material made

from a polymer matrix reinforced with fibres. The fibres are usually glass, carbon, aramid
or basalt. Rarely, other fibers, such as paper, wood or asbestos have been used. The polymer
is usually an epoxy, a vinylester, or a polyester thermosetting plastic, although phenol
formaldehyde resins are still in use. FRPs are commonly used in the aerospace, automotive,
marine, and construction industries. They are commonly found in ballistic armor as well.
2.1 Glass fibre

Glass-Fiber Reinforced Plastics or Polymer (GFRP) uses textile grade glass-fibers. Textile
glass-fibres begin as varying combinations of SiO2, Al2O3, B2O3, CaO, or MgO in powder
form. These mixtures are then heated, through direct melting, to temperatures around 1300°C
37

after which dies are used to extrude filaments of glass-fiber in diameters ranging from 9 to
17 µm. GFRPs are strongest and most resistive to deforming forces when the polymer fibers
are parallel to the force being exerted, and are weakest when the fibers are perpendicular.
2.2 Carbon-fiber
Carbon-fibers are created when Polyacrylonitrile (PAN)-fibres pitch resins, or Rayon are
carbonized (through oxidation and thermal pyrolysis) at high temperatures. Through further
processes of graphitizing or stretching of the fibers, strength or elasticity can be enhanced
respectively. Carbon-fibers are manufactured in diameters analogous to glass-fibers with
diameters ranging from 4 to 17 µm.
2.3 Aramid-fiber
Aramid-fibres are most commonly known as Kevlar, Nomex and Technora. Aramids are
generally prepared by the reaction between an amine group and a carboxylic acid halide
group (aramid), and commonly this occurs when an aromatic polyamide is spun from a
liquid concentration of sulfuric acid into a crystallized fiber. Fibers are then spun into larger
threads in order to weave into large ropes or woven fabrics.
3 LITERATURE REVIEW
Arduini and Nanni (1997) studied pre-cracked, RC beam specimens strengthened by Carbon
Fiber-Reinforced Polymer (CFRP) which were analyzed for different parameters including
two CFRP material systems, two concrete surface preparations, two RC cross sections, and
number and location of CFRP plies. It was observed that the effect of CFRP strengthening
was considerable, but the effect of some of the tested variables was modest. An analytical
model created could simulate the load-deflection behavior as well as the failure mode of
the pre-cracked RC specimens. Different failure mechanisms from ductile to brittle were
simulated and verified.
Norris et al. (1997) experimentally and analytically studied the behavior of damaged or
under strength concrete beams retrofitted with thin CFRP sheets, epoxy bonded to enhance
their flexural and shear strengths. The effect of CFRP sheets on strength and stiffness of the
beams for various orientations of the fibers, with respect to the axis of the beam, was also
considered. It was observed that the magnitude of the increase in strength and stiffness, and
mode of failure were related to the direction of reinforcing fibers.
GangaRao and Vijay (1995) experimentally investigated RC beams strengthened with car-
bon fiber wraps to evaluate the enhancement in flexural strength. This was compared to
identical concrete beams strengthened with bonded steel plates. Parameters like strength,
stiffness, compositeness between wrap and concrete, and associated failure modes, were also
evaluated.
Papakonstantinou et al. (2001) experimentally studied the effects of GFRP composite
rehabilitation systems on the fatigue performance of RC beams. The results indicated that
the fatigue life of RC beams with the given geometry and subjected to the same cyclic load
could be significantly increased through the use of externally bonded GFRP composite
sheets. It was found that the beams failed primarily due to fatigue of the steel reinforcement.
Debonding of the GFRP composite sheet was a secondary mechanism in the strengthened
beams.
Kumutha et al. (2007) experimentally investigated the behavior of axially loaded rectangular
columns that had been strengthened with a GFRP wrap. Specimens with different aspect
ratios wrapped with zero, one, and two layers of GFRP were investigated. The test results
showed a linear relationship existed between the strength of confined concrete and lateral
confining pressure provided by the FRP.
38

Al-Salloum (2007) studied the influence of edge sharpness and found that for CFRP wrap-
ping, the best performance was that of the circular specimens. The performance of the square
columns was found to improve by rounding the corners. An analytical model was also devel-
oped and the predicted value of strength, by the analytical solution was found to be in agree-
ment with the experimental value. This was attributed to the fact that jacketing provides a
uniform confining stress around the circular concrete core.
Wang and Hsu (2008) observed that the GFRP jackets were effective in preventing pre-
mature buckling of the longitudinal reinforcing bars in the columns. They conducted experi-
ments on square and rectangular columns wrapped with multiple layers of GFRP. Columns
jacketed with two layers of GFRP failed by debonding and those with six layers, failed by
splitting of the jacket at the rounded corners of the columns. Both failure types occurred at
large axial strain levels. Closed form equations were proposed for calculating the axial load
strength of confined columns based on the experimental results.
Wang and Wu (2008) conducted experiments on confined short square specimens and
found that the strength gain of CFRP confined concrete columns was in direct proportion
to the corner radius ratio. For higher corner radius, the confinement was effective in increas-
ing the ductility as well. Corner radius ratio should be as large as possible to achieve a better
confinement.
Yalcin et al. (2008) experimentally investigated seismic retrofitting of RC columns. The test
results proved that retrofitting by welding of dowel rebars with longitudinal rebars and then
wrapping with CFRP sheets, enhanced its ductility and energy dissipation capacity.
Wu and Wei (2010) conducted an experimental study into the influence of aspect ratio of
specimens on the number of layers in CFRP wrapping. It was found that CFRP wrapping
was more effective for square than for rectangular sections. As the aspect ratio increases from
one to two, the strength gain decreases until it becomes insignificant at an aspect ratio of
greater than two.
Silva (2011) experimentally studied the performance of cylindrical and square RC speci-
mens with a height 750 mm and a cross sectional area of 225 cm². He found that the improve-
ment of axial capacity and ductility either from jackets of Aramid Fiber Reinforced Polymer
(AFRP) or CFRP, was almost equal for cylindrical columns. Columns of the square section
with sharp corners and CFRP jackets showed no improvement in capacity or ductility. For
AFRP jacketing, there was an improvement of load capacity, but no significant improvement
on ductility.
Ranolia et al. (2012) carried out tests to evaluate the influence of the number of wraps, and
position of wrap, on concrete compressive strength. Cylinders of a height of 300 mm and
diameter of 150 mm, using M20 and M40 grade concrete, were cast and tested with different
patterns of FRP wraps, such as one layer, two-layer full wrap, 100 mm center wrap and two
strips wrap. Better confinement was observed when concrete cylinders were fully confined
with GFRP than partially confined. The failure of the specimens took place in the middle
half region of the specimen when full confinement was provided, and at the top and bottom
portion in the case of partially confined specimens.
Parikh and Modhera (2012) studied the application of GFRP on preloaded retrofitted
beams for improving flexural strength of a beam. The parameters were number of layers of
GFRP, preload and arrangement of GFRP sheets. A new effective arrangement of sheets
was compared against the traditional and found that it exhibited better performance and
mode of failure.
Raval and Dave (2013) carried out investigations on GFRP strengthening for RC columns
of circular, square and rectangular shapes. It was found that GFRP wrapping was more
effective for circular columns than square and rectangular columns. This behavior can be
attributed to the variation of confining pressure distribution for rectangular columns from a
maximum at the corners to a minimum in between.
Sharma et al. (2013) investigated the behavior of GFRP wrapped small scale square RC
columns with varying corner radii and suggested a 149% increment in failure load for a
25 mm corner radius with number of layers of wrapping.
39

Widiarsa and Hadi (2013) studied the performance of square RC columns wrapped with
CFRP subjected to eccentric loading. CFRP wrapping enhanced the performance of the
columns by postponing the rupture of the concrete enhancing reinforcement and increasing
column ductility. Wrapping with a minimum of three layers was suggested in order to achieve
significant results.
Yaquab and Bailey (2011) investigated the axial capacity of post-heated circular reinforced
concrete columns repaired with GFRP and CFRP. The tests comprised of columns which
were un-heated, post-heated, post-heated seriously spalled and repaired with mortar and
post-heated and wrapped with FRP. It was found that the strength of the post-heated column
repaired with a GFRP jacket was enhanced by 29%, more than the original strength of un-
heated columns and 122% higher than post-heated columns.
Singh et al. (2014) experimentally studied the effect on strength of beam-column joints ini-
tially stressed to three different stress levels and retrofitted with CFRP jackets under static load-
ing conditions. CFRP sheets in an L-shape design and at 45° orientations to the joint in two
layers were used. The results showed an improvement in the ultimate load carrying capacity as
well as yield load. The use of CFRP jacketing for retrofitting of initially stressed beam-column
joints was found to help regain full strength even if stressed up to 85% of the ultimate load.
CFRP retrofitting of the beam-column joint, shifted the failure of the joint from the column
portion to the beam portion which, in turn, prevented progressive collapse of the structure.
Alsayed et al. (2013) studied the compression behavior of CFRP strengthened wall-like
rectangular RC columns. Strengthening of wall-like-columns with CFRP wrapping, after
modifying the cross section to an elliptical shape, appeared to be an effective method espe-
cially in situations where the ductility requirement was high. Failure of the strengthened
column was due to the crushing of the concrete and buckling of the longitudinal rebars after
yielding. No yielding of transverse rebars was observed due to the confinement provided.
Hay (2014) conducted experimental studies to investigate the overall behavior of RC square
columns, with poor concrete, at the upper part and which were partially strengthened with
CFRP. It was found significant to wrap the poor part only using one layer of CFRP. Increas-
ing the jacket height provided a higher ductility for wrapped columns.
Tahsiri et al. (2015) compared retrofitting of RC beams using FRP and RC jacketing
under three-point loading. It was found that the RC jacketed beams exhibited a flexural fail-
ure identical to the reference beams. Whereas, the FRP strengthened beams failed in a rela-
tively small deflection due to the premature FRP debonding. But, both techniques showed an
increase in strength and energy dissipation capacity. RC jacketed specimens failed in a more
ductile manner than the FRP jacketed ones. The experimental results were validated with an
analytical approach.
Banjara and Ramanjaneyulu (2017) experimentally investigated shear deficient and GFRP
strengthened RC beams under two-point monotonic loading. Detailed 3D non-linear finite
element analysis with perfect bonding, as well as with cohesive zone modeling was under-
taken to simulate the behavior of shear deficient beams. The load-deflection behavior, failure
loads and crack patterns obtained from numerical simulations were validated with that of the
experimental investigations. In all GFRP strengthened beams, mode of failure changed from
shear to flexural failure and showed great improvement in the ductile behavior.
4 CONCLUSIONS
FRP is used most commonly in the repair and reconstruction of damaged or deteriorating
structures. Seismic protection and changes in codal provisions has established the necessity
for strengthening thereby enhancing the performance of structural elements of a building.
Speed of construction, lightness of weight, resistance to corrosion and high strength has
proved FRP to be an effective material for jacketing. But most works are carried out on
columns with small sizes, plain concrete. More investigations are to be carried out on large
reinforced sections. The impact of seismic behavior on damaged structural elements that have
been jacketed partially or fully, are also to be investigated.
40

REFERENCES
Al-Salloum Y.A. (2007). Influence of edge sharpness on the strength of square concrete columns con-
fined with FRP composite laminates. Composites, 38, (Pt B), 640–650.
Alsayed S.H., Almusallam T.H., Ibrahim S.M., Al-Hazmi N.M., Al-Salloum Y.A. & Abbas H. (2013).
Experimental and numerical investigation for compression response of CFRP strengthened shape
modified wall-like RC column. Construction and Building Materials, 63, 72–80.
Arduini, M. & Nanni, A. (1997). Behavior of pre-cracked RC beams strengthened with carbon FRP
sheets. Journal of Composites for Construction, 1(2), 63–70.
Banjara, N.K. & Ramanjaneyulu, K. (2017). Experimental and numerical investigations on the
performance evaluation of shear deficient and GFRP strengthened reinforced concrete beams.
Construction and Building Materials, 137, 520–534.
GangaRao, H.V.S. & Vijay, P.V. (1995). Bending behavior of concrete beams wrapped with carbon
fabric. Journal of Structural Engineering, 124(1), 3–10.
Hay, A.S.A. (2014). Partial strengthening of R.C square columns using CFRP. Housing and Building
National Research Center Journal.
Kumutha R., Vaidyanathan R. & Palanichamy M.S. (2007). Behaviour of reinforced concrete rectangular
columns strengthened using GFRP. Cement & Concrete Composites, 29, 609–615.
Norris, T., Saadatmanesh, H. & Ehsani, M.R. (1997). Shear and flexural strengthening of RC beams
with carbon fiber sheets. Journal of Structural Engineering. 903–911.
Papakonstantinou, C.G., Petrou, M.F. & Harries, K.A. (2001). Fatigue behavior of RC beams
strengthened with GFRP sheets. Journal of Composites for Construction, 5(4), 246–253.
Parikh, K. & Modhera, C.D. (2012). Application of GFRP on preloaded retrofitted beam for
enhancement in flexural strength. International Journal of Civil and Structural Engineering, 2(4),
1070–1080.
Ranolia, K.V., Thakkarb, B.K. & Rathodb, J.D. (2012). Effect of different patterns and cracking in FRP
wrapping on compressive strength of confined concrete. Procedia Engineering, 51, 169–175.
Raval, R.P. & Dave, U.V. (2013). Behaviour of GFRP wrapped RC columns of different shapes. Proce-
dia Engineering, 51, 240–249.
Sharma, S.S., Dave, U.V. & Solanki, H. (2013). FRP wrapping for RC columns with varying corner
radii. Proceedia Engineering, 51, 220–229.
Silva, M.A.G. (2011). Behaviour of square and circular columns strengthened with aramidic or carbon
fibers. Construction and Building Materials, 25, 3222–3228.
Singh, V., Bansal, P.P., Kumar, M. & Kaushik, S.K. (2014). Experimental studies on strength and
ductility of CFRP jacketed reinforced concrete beam-column joints. Construction and Building Mate-
rials, 55, 194–201.
Tahsiri, H., Sedehi, H.T.O., Khaloo, A. & Raisi, E.M. (2015). Experimental study of RC jacketed and
CFRP strengthened RC beams. Construction and Building Materials, 95, 476–485.
Wang, L.M. & Wu, Y.F. (2008). Effect of corner radius on the performance of CFRP confined square
concrete columns test. Engineering Structures, 30, 493–505.
Wang, Y.C. Hsu, K. (2008). Design of FRP-wrapped reinforced concrete columns for enhancing axial
load carrying capacity. Composite structures, 82, 132–139.
Widiarsa, I.B.R. & Hadi, M.N.S. (2013). Performance of CFRP wrapped square reinforced concrete
columns subjected to eccentric loading. Proceedia Engineering, 54, 365–376.
Wu, Y.F. & Wei, Y.Y. (2010). Effect of cross-sectional aspect ratio on the strength of CFRP confined
rectangular concrete columns. Engineering Structures, 32, 32–45.
Yalcin, C. Kaya, O. & Sinangil, M. (2008). Seismic retrofitting of RC columns having plain rebars using
CFRP sheets for improved strength and ductility. Construction and Building Materials, 22, 295–307.
Yaqub, M. & Bailey, C.G. (2011). Repair of fire damaged circular reinforced concrete columns with
FRP composite. Construction and Building Materials, 25, 359–370.
41

Dynamic soil parameter identification using measured vibration

response data of machine foundations
Liya Jerard
Government Engineering College, Thrissur, Kerala, India
S. Arun
Department of Civil Engineering, Government Engineering College, Thrissur, Kerala, India
ABSTRACT: Traditional methods of dynamic soil parameter estimation involve either

experimentations or analytical investigations using measured ground motions. A novel strategy
for the estimation of dynamic soil properties is attempted in this study. The identification of
dynamic soil parameters is posed as an inverse problem combining measured foundation
response data and numerical simulation. The approach consisted of an optimization problem
where the objective was to minimize the error between the measured foundation response
and numerical simulation results. The efficacy of the approach was demonstrated through
numerical experimentations involving modeling of an impact hammer foundation.
1 INTRODUCTION
Dynamic soil properties govern the behavior of soil subjected to dynamic loads. Dynamic
soil parameters can be classified into low strain properties (dynamic Young’s modulus and
shear modulus, Poisson’s ratio, soil damping etc.) and properties mobilized at larger strains.
Estimation of dynamic soil parameters finds major applications in the estimation of response
of machine foundations, dynamic bearing capacity of soil, and soil-structure interaction
effects during propagation of stress waves generated during earthquakes.
For the estimation of in-situ dynamic soil properties, various tests such as vertical block
resonance test, cyclic plate load test, and wave propagation tests are currently used (IS 5249,
1992). Such methods are laborious and require expensive instrumentation.
To simplify the dynamic soil property estimation procedure, researchers in the past have
tried an inverse problem solution approach. Ikemoto et al. (2000) used measured ground
motions during earthquakes for the estimation of dynamic soil parameters. The methodol-
ogy used a combination of the S-wave multiple refection theory and an optimal search tool
based on genetic algorithms. Glaser and Baise (2000) used the time series analysis technique,
the Auto Regressive Moving Average (ARMA) model for the identification of soil properties
from ground motion history. Garsia and Romo (2004) used a neural network approach for
identifying dynamic soil properties from earthquake records.
A time domain approach, involving the use of an extended Kalman filter for identification
of soil stiffness and damping, was proposed by Mikami and Sawada (2004). The methodology
was successfully demonstrated over a soil-structure system modeled using three degrees of
freedom.
This paper presents a novel attempt for in-situ identification of dynamic soil stiffness and
damping parameters. The methodology involves use of measured dynamic responses from
the foundation, numerical models for simulating the dynamic response of a soil-foundation
system and an optimal search tool for the identification of soil parameters. The methodology
is demonstrated through numerical studies on the model of a 16 T impact hammer at the
forging facilities of Steel Industries Forging Limited (SIFL) in Athani, Thrissur.
43

2 PROPOSED METHODOLOGY
The main objective of the work was to explore the potential of identifying the dynamic soil
properties from the measured time history of dynamic responses of the foundation. For this,
the numerical model of a coupled soil-foundation system was utilized. The parameters of the
numerical model were treated as unknown variables and the optimal value of the same were
searched for the condition that the error between the measured foundation response history
and simulated response history from the numerical model was a minimum.
The numerical evaluation of dynamic response of the soil-foundation system and optimal
search algorithms formed the major computational part of the methodology. For this study,
Newmark’s time stepping method (Chopra, 2015) was employed for the numerical integra-
tion of the coupled equations of motion of the soil-foundation model. Newmark’s average
acceleration scheme, which has unconditional numerical stability, was utilized.
The soil dynamic parameters, the dynamic soil stiffness (which is a function of dynamic
modulus of elasticity) and soil damping ratio, was kept as the unknown parameters to be
identified. The optimal values of these parameters were searched from a set representing the
practical range of these parameters.
Conventional methods of soil parameter identification involve the use of gradient based
search algorithms, but such methods suffer from the problem of converging to a local min-
imum. Hence for the present study, genetic algorithms were used for the optimal search.
Genetic algorithms are a computerized search and optimization algorithms are based on the
mechanics of natural genetics and natural selection. They are good at taking larger, poten-
tially huge, search spaces and navigating them looking for optimal combinations of things
and solutions (Rajasekaran and Vijayalekshmi, 2003).
Genetic algorithms proceed to the global optimum through the following major steps:
1. Generation of random population. Solving a problem means looking for a best solution;
the space where all possible solutions may be present is called a ‘search space’. For this
study, the search space was generated by specifying lower and upper limits of unknown
variables (stiffness of soil or damping of soil), with a uniform distribution between the
limits. A random population was initially generated from the defined search space.
2. Evaluation of fitness for each organism. As per Darwin’s evolution theory of “survival
of the fittest”, the best ones should survive and create new offspring. The fitness of each
organism (i.e., each individual of the initial random population) needs to be evaluated.
A fitness function derived from the objective function was used for this purpose.
Genetic algorithms are suitable for maximization problems. Hence minimization
problems need to be converted to maximization problems by some suitable transforma-
tion (Rajasekaran and Vijayalekshmi, 2003). Generally, this transformation is achieved
through a suitable inversion of the objective function. Alternatively, an exponential map-
ping procedure was found to be more effective (Gregory, 1991) and was utilized in this
study. The exponential mapping for a minimization problem used is defined by:
F ( x ) = exp( − β × Objs worstObj ) (1)
where β is the Selection Pressure, defined as the ratio between the probability that the
most fit member of the population is selected as a parent to the probability that an average
member is selected as a parent. Its value is taken as 8.
Objs is the Objective Function Value of Organism (sample from a selected random
population) whose fitness is evaluated. Worst Objective Function Value (worstObj) is, the
organism with maximum error in the initial random population.
3. Selection of models into new population. Samples are selected from the initial random
population to cross over and produce the best offspring based on its fitness value. Selec-
tion is completed by any of these three procedures: roulette wheel selection; tourna-
ment selection; or random selection. Roulette wheel selection procedure was used in this
study.
44

4. Crossover and mutation. The crossover operator is applied to create new offspring. It is a
recombination operator, which proceeds in three steps. First, the reproduction operator
selects, at random, a pair of two individual strings for mating, then a cross-site is selected
at random along the string length and the position values are swapped between two strings
following the cross-site. Mutation follows crossover. The mutation operator introduces new
genetic structures in the population by randomly modifying some of its building blocks.
The following error norms were used to compare the measured response history with that
from the numerical simulation, and the most effective among the two was identified.
1. Mean Square Error (MSE) is defined as:
1 n
∑ (X i − xi )
2
MSE = (2)
n 1
where X is the Measured Response, x is the Simulated Response from the numerical model
and n is Overall Number of Time Steps used in Analysis or Sampling Interval of the
Measured Response.
2. Absolute difference of L2 norms of data are:
err 2 = XX T − xxT (3)
The first norm is more stringent as it estimates the error point wise, rather than the second
norm, which estimates the error associated with data in a mean sense. The optimization
procedure involved finding the global minimum of this error norm for the practical range
of soil parameters to be identified. The overall methodology used is presented in the form
of a flowchart in Figure 1.
A MATLAB script based on the above flowchart was developed for performing numerical
studies.
Figure 1. Overall methodology.
45

3 CASE STUDIES
The methodology was demonstrated for the case of the identification of dynamic soil para
meters for a 16 T impact hammer foundation located in forging facilities of Steel Industrial
Forging Limited (SIFL) in Athani, Thrissur. An impact hammer foundation is a structure
used to receive and transfer both static and dynamic loads imposed during the operation of
the machine (Figure 2).
The following details relate to the existing SIFL 16 T hammer foundation:
1. Anvil mass = 325 T.
2. Foundation mass = 1225.32 T (based on the foundation drawings).
3. Stiffness of neoprene pad between anvil mass and foundation = 9 × 106 kN/m (based on
the elastic modulus of the material used as specified in the drawing and the dimensions of
the pad).
4. Stiffness of soil at foundation level = 3.93 × 106 kN/m (based on the bearing capacity of
soil at the foundation level of 8 m).
The soil-foundation system was modeled as a two degree of freedom system and the equa-
tion of motion for the same was developed as:
M 1 0   x1   C1 −C1   x1   K1 − K1   x1  0 

 0 M  x  +  −C C + C  x  +  − K K + K  x  = 0  (4)
 2 2  1 1 2 2  1 1 2 2  
M1 is Mass of the Anvil, K1 is Stiffness of the Elastic Pad, M2 is Mass of the Foundation
Block, K2 is Soil Stiffness, C1 is Damping (Rajasekaran and Vijayalekshmi, 2003). Associated
with Elastic Pad, and C2 is Soil Damping.
The right hand side of Equation 2 was zero as the problem was of initial value type, with
the initial velocity of impact being computed from the energy transferred and fall of impact
hammer (V = 6 m/s).
The soil stiffness, K2, can be related to the Coefficient of Uniform Compression (Cu) and
Area of Foundation Block (Ab) through the relation (IS 5249: 1992).
K 2 = Cu Ab (5)
The soil modulus of elasticity and coefficient of uniform compression holds the following
relationship (IS 5249: 1992):
Figure 2. Impact hammer foundation (IS 5249: 1992).
46

1.13 E
Cu = (6)
(1 − µ 2 ) Ab
where Cu is the coefficient of uniform compression in kN/m3; µ the Poisson’s Ratio of Soil
and E is the Soil Elastic Modulus.
Finally, the Soil Damping Ratio (ζ) can be related to Damping Coefficient (C2) through
the foundation mass and natural frequency of vertical vibrations.
Since the objective of the present study was to understand the efficacy of the proposed
identification scheme, observation records of foundation response were synthetically gen-
erated by numerically computing the response of the soil-foundation system using known
values of soil parameters to be identified (soil stiffness and damping).
3.1 Case study 1: Single parameter identification: Identification of soil stiffness

Soil stiffness was treated as the unknown parameter to be identified. Soil and neoprene pad-
ding damping were not considered in the analysis. Initial population (sample space) for the soil
stiffness was chosen in the practical range of 3 × 106 kN/m to 14.5 × 106 kN/m (Barkan, 1962).
The results of simulation are presented in Table 1. Simulation using MSE as an error
estimate and objective function had a faster convergence (7 iterations) compared to the one
using err2 as an error estimate (12 iterations). Hence it is concluded that the mean square
error estimate was a better measure for defining the objective function in identification com-
pared to err2. The response history of foundation from the model corresponding to the con-
verged/identified soil stiffness (3.93 × 106 kN/m) compared well with the synthetic response
(Figure 3). From the identified soil stiffness, the modulus of elasticity can be computed using
Equations 3 and 4 as 241.44 × 106 N/m2.
Table 1. Results for identification of stiffness alone.
Error norm MSE Err2
Error in parameter estimation 0 0

Number of iterations to convergence 7 12
Figure 3. Comparison of synthetic response and converged response from simulations (using identi-
fied parameter).
47

Figure 4. Comparison of synthetic response and response from simulations corresponding to identi-
fied parameters.
3.2 Case study 2: Multi-parameter (Stiffness and damping of soil) identification

The same algorithm was utilized here for the estimation of soil damping and stiffness.
Synthetic data representing the measured response of foundation was generated for the fol-
lowing set of soil and foundation parameters:
Anvil Mass is M1 = 325 T, Foundation Mass is M2 = 1225.32 T, Neoprene Pad Stiffness
is K1 = 9 × 109 N/m, Soil Stiffness is K2 = 3.93 × 109 N/m, Neoprene Pad Damping is
C1 = 7.572 × 106 N-s/m, and Soil Damping is C2 = 15.96 × 106 N-s/m. Minimization of
mean square error was set as the objective function. For sampling the soil stiffness and soil
damping in the optimization scheme, the practical limits of the same were set as 3 × 109 N/m
to 14.641 × 109 N/m, and 1.467 × 106 Ns/m to 18.189 × 106 Ns/m, respectively (Barkan, 1962).
Keeping minimization of the mean square error as the objective function, the proposed
algorithm could identify both soil stiffness and damping with error less than 2% in 40
iterations. The time history of foundation response using the converged/identified values of
soil stiffness and damping parameters is shown in Figure 4.
4 SUMMARY AND CONCLUSIONS
This study explored a strategy for the estimation of in-situ dynamic soil properties from
dynamic response measurements on foundation structure rather than direct measurements
of soil response. The methodology combined a numerical time integration procedure (for the
solution of coupled equations of motion involving soil mass and foundation structure), and
an optimization algorithm (for identification of the optimal value of soil parameters) within
a system identification framework. The efficacy of the methodology was demonstrated
through case studies on impact hammer foundation. The following conclusions were derived
from the present study:
Dynamic soil parameters can be estimated by exploiting the potential soil-structure inter-
action using system identification techniques.
Newmark’s method and genetic algorithm proves to be an excellent numerical tool in this
venture.
Mean square error was found to be a better estimate for defining the objective function in
the optimization scheme for the identification of soil parameters.
The proposed methodology is able to identify the dynamic soil parameters of soil stiffness
(elastic modulus) and soil damping with 2% error.
48

The proposed methodology has been successful in identifying the dynamic soil parameters
for the cases considered in this paper. But it is acknowledged that the success of the identifi-
cation methodology will be higher for soil-foundation systems with a higher level of interac-
tion between the two. Further studies in this regard are required for quantifying the effect
of interaction in soil parameter identification. For this study, since the focus of work was
on identification of soil parameters for an impact hammer foundation, the vertical transla-
tion degree of freedom of the foundation alone was included in the soil-foundation model.
This study could further be extended by including other translation and rocking degrees of
freedom.
ACKNOWLEDGEMENTS
This work was carried out as part of an undergraduate project for Liya Jerard. The manage-
ment of SIFL is thanked for giving access to their forging facilities in Athani, Thrissur. The
contribution of other batch mates’ (Panchami, Sameera, Sabareesh and Varghese) is also
acknowledged.
REFERENCES
Barkan, D.D. (1962). Dynamics of bases and foundations. New York, NY: McGraw-Hill.
Bureau of Indian Standards. (1980). Code of practice for design and construction of machine foundations
IS 2974 Part II: 1980). New Delhi, India: Author.
Bureau of Indian Standards. (1992). Determination of dynamic properties of soil—Method of test IS
5249: 1992. New Delhi, India.
Chopra, A.K. (2015). Dynamics of structures (3rd ed.). New Delhi, India: Pearson.
Garsia, S.R. & Romo, M.P. (2004). Dynamic soil property identification using earthquake records:
A neural network approximation. In, Proceedings of the 13th World Conference on Earthquake
Engineering. Vancouver, Canada.
Glaser, S.D. & Baise, L.G. (2000). System identification estimation of soil properties at Lotung site. Soil
Dynamics and Earthquake Engineering, 19, 521–531.
Gregory, J.E.R. (1991). Foundations of genetic algorithms. San Mateo, California: Morgan Kaufmann
Publishers.
Ikemoto, T., Miyojima, M. & Kitaura, M. (2000). Inverse analysis of soil parameters using accelera-
tion records. In Proceedings of the 12th World Conference on Earthquake Engineering. Auckland,
New Zealand.
Mikami, A. & Sawada, T. (2004, August). Time-domain identification system of dynamic soil-structure
interaction, Proceedings of the 13th World Conference on Earthquake Engineering. Vancouver,
Canada.
Rajasekaran, S. & Vijayalakshmi Pai, G.A. (2003). Neural networks, fuzzy logic and genetic algorithm.
New Delhi, PHI Learning Pvt. Limited.
49

Pushover analysis of GFRG-OGS building systems
Krishna S.R. Gouri, Devdas Menon & Prasad A. Meher

Indian Institute of Technology, Madras, Chennai, India
ABSTRACT: Glass Fibre Reinforced Gypsum (GFRG) panels have been in use in our
country since a decade. These panels, with the advantages of rapidity, sustainability and
affordability, are best suited to meet the mass housing requirements of India. However, the
present design of GFRG buildings needs the walls to start from the foundation itself as the
walls are load bearing. This denies the provision of ground storey parking which is a feature
of high demand in multi storey constructions especially in urban areas, where the land is
scarce. The proposed solution to cater this issue is to raise the GFRG building system over a
frame structure in ground storey comprising of Reinforced Concrete (RC) columns, beams
and slabs. The present study evaluates the behaviour of GFRG Open Ground Storey (OGS)
structure subjected to gravity and lateral loads using pushover analysis.
1 INTRODUCTION
1.1 General
Glass Fibre Reinforced Gypsum (GFRG) panels have been used for the construction of
GFRG buildings in India for more than a decade. They are suitable for rapid affordable
mass-scale building construction. These are load bearing hollow panels manufactured by
means of a special calcination process converting raw gypsum, obtained as waste product
from the fertilizer industries into calcined gypsum plaster, which is then reinforced with glass
fibres along with the addition of special additives. The addition of 300–350 mm long glass
fibres, randomly spread, contributes tensile strength to the brittle gypsum plaster, and the
inclusion of additives offers moisture resistance and load-bearing capability. Figure 1 shows
the typical cross section of the panel.
GFRG panel, as a structural member, is capable of resisting axial load, lateral shear &
in-plane bending and out-of-plane bending. In Australia, where the panels were invented
in 1990, these were used as load bearing walls in low-rise as well as multi-storey build-
ings (up to 9 stories). Subsequently, the technology was adopted in various other countries
such as China and India, based on research works carried out in these countries. The stud-
ies in India carried out at the Structural Engineering Research Centre (SERC), Chennai
and Indian Institute of Technology Madras (IITM) helped in gaining acceptability in the
Figure 1. Cross section of GFRG panel.
51

Figure 2. Collapse of open ground storey building during 2001 Bhuj earthquake.
country. Studies at IITM helped in extending the use of the panels for the construction of
slabs and stair cases as well as shear walls capable of resisting seismic loads. Infilling of
concrete inside the cavities of the panel enhances the axial load bearing capacity more than
9 times, and enhances the lateral load carrying capacity tremendously with the help of rein-
forcement. Hence, for all major structural applications, these panels are used in combination
with reinforced concrete. Construction using GFRG panels offers advantages of afford-
ability, faster construction, eco-friendliness, increased carpet area, reduction in weight of
structure, etc. The GFRG buildings constructed so far have the walls starting from the foun-
dation itself.
The current demand from the housing sector for the urban community in India is the pro-
vision of parking space for vehicles at the ground or basement storey of the building, referred
to as open ground storey (OGS) building system.
OGS for conventional multi-storeyed framed structure demands either a dynamic analysis
of the building or an enhancement factor (varies depending on different codes) for the forces
(obtained from analysis) in the ground storey columns and beams, in order to avoid the for-
mation of plastic hinges in the columns, which will result in sudden collapse of the structure
(Fig. 2). Recently, a recommendation of minimum wall plan density was also introduced in
IS 1893 Part 1:2016.
1.2 State of the art

1.2.1 Studies on GFRG panels
Though the panels were invented in 1990 in Australia, major studies on the panels were
reported only in the 2000s. Wu (2002) pioneered the research and development of various
aspects of GFRG panels. Further studies were also carried out in China, Malaysia and India
focussing on experimental and theoretical aspects.
Preliminary tests to determine the physical properties of GFRG were carried out by Wu
(2002). These tests comprised of the evaluation of material properties such as compressive
strength, tensile strength, flexural strength, modulus of elasticity, water absorption, thermal
expansion coefficient etc. Similar tests were also carried out in India (SERC, Chennai and
IIT Madras). Based on these test results, standard values were proposed to assess the quality
of the panels.
Strength aspects of GFRG panels were studied in various countries across the world.
Studies on panels subjected to axial and eccentric compression, monotonic and cyclic lateral
loads were evaluated with emphasis on peak load, failure mode and load-deformation behav-
iour. Compressive strengths were evaluated by changing different variables such as width of
panel, number of cavities infilled, boundary condition, number of reinforcements in cavities
etc. Unfilled panels failed by crushing of plaster and the concrete filled panels by buckling.
52

Also it was reported that the grade of concrete infill and the presence of reinforcement has
no effect on the compressive strength of panels. The cyclic load studies on GFRG panels
helped in quantifying energy dissipation and ductility. Some additional studies were also
done under cyclic loading for evaluating the influence of continuity of reinforcement and
presence of opening.
Walls with width greater than or equal to 2 m showed predominant shear behaviour. The
behaviour of such walls were characterised by the 45° shear cracks developed before the peak
load (at around 70% of peak load for concrete filled panels) and visible longitudinal shear
cracks at peak load. The longitudinal cracks were formed as a result of loss of bond between
GFRG and concrete infill at peak load which attracted more shear load in GFRG and its
failure. It can be stated that “the gypsum panel acted as a “link” between the concrete cores
that restricted free relative movements between the concrete cores” (Wu & Dare, 2004).
Full scale building models were also tested in order to study their response and failure
mechanism. In one of the tests, a five storeyed building was subjected to quasi static lateral
load, for which the bending type deformation was observed initially as the GFRG panels and
the concrete infill were together resisting the load. After cracking the bond between GFRG
and infill was lost and the performance of the panel reduced and failed gradually with a
shear type deformation. GFRG buildings were also subjected shake table loading to simulate
ground motion. Buildings with different plan and infill configurations were tested and only
minimum structural damage was observed in the form of hairline cracks.
Recent studies in IIT M on GFRG wall panels revealed that the percentage of reinforce-
ment in each cavities does not have much effect on the lateral load carrying capacity of walls
with width greater than 2 m (Philip, 2017). Nevertheless, the post peak behaviour seems to
get affected.
1.3 Studies on OGS buildings

When there is significant change in the distribution of mass, geometry, stiffness or strength
along the height, then the structure is called a vertically irregular structure. OGS buildings
belong to this category of vertical irregularity. In the case of OGS buildings, the stiffness and
strength of the ground storey are much less compared to the upper storeys.
1.3.1 Code recommendations

The design provisions in various codes and literature for buildings with vertical irregularity
have been studied. As per IS 1893 Part-1:2002, the OGS system falls under the weak-storey
and soft-storey (sometimes extremely soft) types of vertical irregularity. Special provisions
are considered in the code for the design of these irregular structures. A dynamic analysis is
suggested including effects of infill walls. In the absence of a proper dynamic analysis, the
code permits an equivalent static bare frame (without infill stiffness) analysis, but imposes a
magnification factor of 2.5 on the bending moment and shear force demand on the columns
of open ground storey. In the recent revision of code (IS 1893 Part-1:2016), a minimum
structural wall plan density of 2% in ground storey is suggested for the design of OGS build-
ings. ASCE 7–10 recommends dynamic analysis for vertically irregular buildings. Other codes
such as EC 8 (2004) recommends alternative formulation for the magnification factor. These
effects applicable to conventional OGS framed structures have been studied in detail by vari-
ous researchers, such as Kaushik (2006), Davis (2009), Jiji (2013), etc.
1.3.2 Composite wall—beam action

When a wall is resting over an extremely rigid base, the load as well as the self-weight of
the wall will be distributed uniformly over the base. In the case of GFRG-OGS buildings,
the walls are supported on a relatively flexible beam of the ground storey frame. Hence the
deflection of the beam, which is more flexible compared to the wall, will cause a loss in con-
tact between them near the mid span. This will eventually lead to a concentration of stresses
near the support where the contact is intact. This is commonly referred to as arch action
arising due to wall—beam interaction (Wood, 1952).
53

The major studies on this topic were mainly concentrated on structural masonry walls
which was the most common case. The studies on this area were initiated by Wood (1952).
He put forward the guidelines for calculating bending moments in wall—beam structures.
He also established the need for a sufficiently strong link between the wall and the beam,
in order to allow the development of the shearing forces required for the complete develop-
ment of the composite action between the wall and its support beam. In 1981, Hendry et al.,
proposed a limit of 0.6 for the wall height to span ratio. This was re-established by Jagadish &
Ramachandra (2000), indicating the flattening of arch when the ratio is less than 0.5.
In short, the behaviour of GFRG-OGS system is different from the conventional RC
framed OGS buildings. Hence the provision of OGS in GFRG buildings calls for a proper
study. This provides the motivation for the present research.
2 DESIGN PHILOSOPHY
The basic design philosophy of the proposed GFRG building system is to avoid the collapse
by storey mechanism due to the formation of plastic hinges in the ground storey. In order to
satisfy this, the columns and beam in the ground storey shall be designed in such a way that
the yielding of frame members will not precede the attainment of peak load by the GFRG
wall in the upper storey. The dissipation of energy will be taken care of by the formation of
vertical cracks in GFRG.
3 MODELLING
The modelling was done in SAP2000. The GFRG wall was modelled as a series of nonlinear
layered shell elements connected with link elements (representing GFRG rib between the
concrete cores). The shell element constituted two layers of GFRG (top and bottom flange),
one layer of M20 concrete (infill inside the cavities) and a reinforcement layer oriented in
90° (12 mm longitudinal reinforcement inside the cavities). Figure 3 shows the details of the
SAP2000 model of a GFRG wall.
The modelling of GFRG wall was validated with the available experimental results (Philip,
2017) as shown in Figure 4. The GFRG wall was connected to an RC plinth beam at the bot-
tom by means of starter bars. The plinth beam was locked on to the strong floor of the labo-
ratory to arrest any sliding or uplift of the beam. Hence the wall was modelled with fixity at
one end. The experimental specimen was subjected to quasi static lateral load at the top level.
The performance of GFRG-OGS building system was compared with that of a conven-
tional RC—OGS framed building in terms of base shear drift demand. The analysis was
Figure 3. Modelling of GFRG wall panel in SAP2000.
54

Figure 4. Validation of modelling of GFRG wall with experimental results.
Figure 5. Cross section details of ground storey frame and GFRG walls.
done for a representative 4 storey 5 bay frame assuming a symmetric plan for the building.
A bay width of 3 m and a storey height of 3 m were considered in this study. Figure 5 shows
the details of cross section of columns and beams in the ground storey and the cross section
of the GFRG panel in the upper storeys.
The sections were chosen based on a study conducted in RC—OGS buildings (Davis,
2009). Hinge properties developed from the moment—rotation curves of these sections were
also assigned to the frame elements. A nonlinear static displacement controlled analysis was
carried for the chosen frame. A uniform lateral load distribution was chosen for the push-
over analysis according to the FEMA 356 recommendation for vertically irregular struc-
tures. Hence a unit load was applied at the top level of each storey with a specified target
displacement.
The arch action in the transfer of gravity load, as stated in the literature, was observed
in the analysis. Figure 6 shows the principal stress distribution in the walls under gravity
load.
It was observed that no hinges were formed in the ground storey frame till a drift level of
0.5%. For the RC—OGS frames in the literature, even with a drift ratio of 0.11%, plastic
hinges were formed, though they were within life safety performance level. This indicates
that the demand in the columns and beams in the GFRG-OGS building system is much
lesser compared to the conventional OGS structures. Figure 7 shows the pushover curve for
GFRG-OGS specimen obtained from SAP2000.
Figure 8 shows the hinges formed in the ground storey frame at a higher drift ratio of 1.6%.
55

Figure 6. Principal stress distribution in GFRG walls under gravity load.
Figure 7. Result of pushover analysis on GFRG-OGS system in SAP 2000.
Figure 8. Hinges in the ground storey frame.
56

4 CONCLUSION
The behaviour of GFRG-OGS building systems are different compared to RC framed OGS
building systems. The performance was characterized by the arch mechanism of gravity
load transfer which resulted in lesser force demand in the ground storey beams. It also
exhibited better performance in terms of delayed formation of plastic hinges in the ground
storey columns. The larger values of drift ratios obtained with the adopted dimensions also
indicate the possibility of reduced member cross sections compared to RC framed OGS
buildings.
ACKNOWLEDGEMENTS
The authors would like to thank Indo-French Centre for the Promotion of Advanced
Research (IFCPAR), India, for their continued support throughout this research work.
REFERENCES
Arlekar J.N., Jain S.K. and Murty C.V.R. 1997, Seismic response of RC frame building with soft storeys,
Proceedings of the CBRI golden jubilee conference on natural hazards in urban habitat, November,
New Delhi: 3–24.
ASCE 7 2005, Minimum design loads for buildings and other structures, American Society of Civil
Engineers, USA.
Burhouse, P. 1969. Composite action between brick panel walls and their supporting beams, Proc. Instn.
civ. Engrs, v. 43, p. 175–94.
Coull A. 1966, Composite action of walls supported on beams, Build. Sci., Vol. 1: 259–270.
FEMA 356 2000, NEHRP recommended provisions for seismic regulations for new buildings and other
structures. Federal Emergency Management Agency, Washington DC, USA.
Green D.R. 1972, Interaction of solid shear walls and their supporting structures, Build. Sci., Vol. 7:
239–248.
IS 1893 (Part 1): 2002, Indian standard criteria for earthquake resistant design of structures—General
provisions and buildings, Bureau of Indian Standards, New Delhi.
IS 1893 (Part 1): 2016, Indian standard criteria for earthquake resistant design of structures—General
provisions and buildings, Bureau of Indian Standards, New Delhi.
Janardhana, M. 2010, Studies on the behavior of glass fiber reinforced gypsum wall panels, PhD Thesis,
Indian Institute of Technology Madras, India.
Jiang X. and Gu Y. 2007, Cyclic behaviour of fibre-reinforced plaster board with core concrete composite
shear walls, Proceedings of 9th Canadian Conference on earthquake engineering, September 26–29,
2007, Ottawa, Ontario, Canada, pp. 1234–1242.
Jiji Anna Varughese 2013, Displacement-based seismic design of RC frame buildings with vertical
irregularities, PhD Thesis, Indian Institute of Technology Madras, India.
Kaushik H.B. 2006, Evaluation of strengthening options for masonry-infilled RC frames with open first
storey, Ph.D. Thesis, Indian Institute of Technology Kanpur, India.
Kuang J.S. and Shubin Li. 2005, Interaction-based design formulas for transfer beams: box foundation
analogy, Practice Periodical on Structural Design and Construction, Vol. 10(2): 127–132.
Liu K., Wu Y.F., and Jiang X.L. 2008, Shear strength of concrete filled glass fiber reinforced gypsum
walls, Materials and Structures, 41(4): 649–662.
Muthumani, K., Lakshmanan N., Gopalakrishnan S., Krishnamurthy T.S., Sivarama Sarma B.,
Balasubramanian K., Gopalakrishnan N., Sathish Kumar K., Bharat Kumar B.H., Sreekala R. and
Avinash S. 2002, Investigation on the behaviour of Gypcrete panels and blocks under static loading,
A report prepared by Structural Engineering Research Center, for M/S Gypcrete Building India (P)
Ltd., Chennai, India: 1–19.
Philip Cherian 2017, Performance evaluation of GFRG building systems, PhD Thesis, Indian Institute
of Technology Madras, India.
Robin Davis P. 2009, Earthquake resistant design of open ground storey RC framed buildings. Ph.D.
Thesis, Indian Institute of Technology Madras, India.
SAP 2000 NL Manual, Integrated software for analysis and design, computers and structures, Inc.,
Berkeley, California, USA.
57

Sreenivasa 2010, Strength and behaviour of glass fiber reinforced gypsum wall panels, PhD Thesis,
Indian Institute of Technology Madras, India.
Wood R.H. 1952, Studies in composite construction, Part I. The composite action of brick panels
supported on reinforced concrete beams, National Building Studies Research, Paper No. 13.
Wu, Y.F. 2009, The structural behavior and design methodology for a new building system consisting of
glass fiber reinforced gypsum panels, Construction and Building Materials, 23(8): 2905–2913.
Wu Y.F., and Dare M.P. 2004, Axial and shear behavior of glass fiber reinforced gypsum wall panels:
Tests, Journal of Composites for Construction, 8(6): 569–578.
Wu Y.F. 2002, A 2002 report in to the physical testing and the development of design guidelines for the
structural application of Rapidwall in building construction, Dare Sutton Clarke Engineers, Adelaide,
South Australia: 4–145.
Wu Y.F. 2004, The effect of longitudinal reinforcement on the cyclic shear behavior of glass fiber
reinforced gypsum wall panels: tests, Engineering Structures, 26(11), 1633–1646.
Yue, J. and Bao P. 2007, Test on glass fibre reinforced gypsum panels partially filled with RC, The Indian
Concrete Journal, February, 2007: 7–14.
Zhao K. and Zhang X. 2008, Full-scale model test research on the performance of a five-storey fibre
plaster board building, Proceedings of the 7th International RILEM Symposium on Fibre Reinforced
Concrete: Design and Applications (BEFIB 2008), 17th–19th September, 2008, Chennai, India.
58

Low-velocity impact behaviour of composite laminates—a review
Alen Austin & R.S. Priyadarsini

ABSTRACT: Fibre-reinforced polymer based products have been increasingly used in

the automobile, aerospace and aircraft industries in view of its high strength and stiffness
properties while being light in weight. They are highly susceptible to impact damage during
its fabrication, manufacture, normal operation and during maintenance stages of its useful
life cycle. Initiation and propagation of the resulting cracks leads to a dramatic loss of the
structural utility of these materials. There has been an abundance of research dealing with
detailed analysis of post-impact strength characteristics of laminated composite materials.
This paper presents a comprehensive review of the various strategies, procedures and
methodologies adopted by researchers to evaluate the influence of low velocity impacts
on carbon-fibre and glass fibre-reinforced epoxy composite specimens. The influence of
the composites and its constituents on impact properties of laminates are discussed and
post-impact performance is assessed in terms of its residual strength. A critical evaluation
of various conditions and parameters that are capable of serving as palliatives to impact
damage behaviour of laminated composites and future scope and directions in this research
field are also discussed.
1 INTRODUCTION
Research and development of innovative materials is an important step in the engineering

of utilities for the service of mankind. Relentless studies in advanced and smarter materials
have resulted in development of newer materials and technologies, including that of high-
performance plastics, at an amazing pace over time. Earliest significant applications of
composite materials can be traced back to World War II wherein composite materials were
fabricated for a wide variety of military-based applications. With time, their applications
were extended to a variety of other industries as automobiles, aircrafts, aerospace shuttles
and modules etc. Hence a comprehensive understanding and in-sight into the behaviour
of these materials became an essential pre-requisite for proper and economical design of
elements based on these materials at lower costs.
A composite material is one which is fabricated by combining 2 or more materials to obtain
products having a good blend of each of its individual constituent materials. This abstract
definition of composite materials serves to include a wide variety of materials that can be
termed as composite materials including both natural and artificial synthetic materials. Fibre
reinforced laminates differ from natural materials in that constituent materials are separate
phases mechanically and have different molecular structures. Fibre-reinforced materials thus
tries to attain the desirable properties of the single individual constituents onto a single layup.
Fibres provides the necessary strength and stiffness while the matrix holds together the fibres
and aids in transfer of loads to fibres. Carbon fibres and glass fibres are among the most
widely used fibres for structural purposes.
This review has been carried out as a part of a research programme investigating the
influence of low velocity impact on the properties of laminated composite plates in view of
its increased applications in automobile and space industries and its growing influence in the
civil engineering industry.
59

2 BACKGROUND AND AREA OF STUDY
2.1 Background
Impact damage resistance of laminated composites assumes significance in the wake of
extensive use of such composites especially in bodies of automobiles, fuselage and wing skin
of aircrafts, skin of satellite modules, the application of which were made possible by the
very high stiffness and strength properties of fibre-reinforced plastics when compared to
other engineering materials for the same weight.
The use of composite materials in the aircraft industry have become so extensive that it now
constitutes a major portion of its structural weight. However, the considerable advantage of
weight savings offered by the use of fibre-reinforced polymers are restricted by the conserva-
tive design philosophy still followed for a safe design of costly fibre-reinforced polymers and
can be attributed to the underestimated design strength of these composite materials mainly
based on the concern of the extent of influence of low-velocity impact on the strength, stiff-
ness and damage tolerance of composite laminates. Out-of-plane impact by foreign objects,
runway debris and dropped tools, is expected to occur during the operation, manufacturing,
maintenance and service of composite components (Farooq, 2014). This type of impact usu-
ally leaves damage that is hardly detectable by visual inspection; referred to as barely visible
impact damage (BVID) which significantly reduce the structural performance of composite
laminates under service loads.
2.2 Target audience

The target audience of this review paper are those who wish to conduct an extensive study
into the impact damage of composite laminates regarding its low-velocity impact behaviour.
The major point to be noted is that during impact testing, no quantification of values during
the entire testing slot are possible as the impacting process gets completed with a fraction
of a second. However, recent studies in this field have made use of high-end instrumenta-
tion to determine the impact force time history in terms of micro-seconds. Therefore, for
a majority of impact-related studies on composites, the only significant outputs obtained
included the parameters of the impact damaged specimens as the final profile of the speci-
men, the final deformation value and its internal damage mapping using non-destructive
techniques. Quantitative inputs canalso be obtained by conducting compression and ten-
sion testing on the impact damaged specimens. Compression after impact testing highlights
the fibre-breaking sequence and progression of the test samples since the process of com-
pression tends to close up the other damage failure modes of laminate specimens mainly
de-laminations. Tension testing on impact damaged specimens gives an idea of the dela-
mination sequence within the tested specimen as on applying tension, the de-laminations
within the specimens open up causing the laminate specimens to break and buckle around
the delaminated portions.
2.3 Precautions for the study

Other important precaution to be taken in this study is the SCALING (up/down) of the
properties of the laminated composites to represent the actual behaviour for design of sub-
components prepared from the laminated specimens. Since properties of composite materi-
als are usually NOT strain-dependent, the scaling of strains from the model adopted for the
study to a larger prototype may be possible but the scaling of damage of the specimens is not
possible in a similar manner.
3 LOW VELOCITY IMPACT
Generally, impacts are categorized into either low or high velocity impact but a clear cut
differentiation of a low velocity impact from a high velocity impact event in terms of the
60

impact velocities is still a point of debate. Many literatures consider an impact velocity of
around or less than 10 m/s as being a low velocity impact event (Richardson and Wisheart,
1996). Differentiation of a low velocity impact from a high velocity impact event can also be
made in terms of the damage the test specimen has undergone—high velocity impact is thus
characterised by penetration induced fibre breakage while low velocity impact is character-
ised by delamination and matrix cracking. In terms of the material response to the impact
to which it is subjected, it can be said that the high-velocity impact response is dominated
by stress wave propagation through the material, in which the structure does not have time
to respond, leading to a localised damage. In low-velocity impact, the dynamic structural
response of the target is of utmost importance as the contact duration is long enough for
the entire structure to respond to the impact and in consequence more energy is absorbed
elastically.
3.1 Importance of drop-weight tests over Charpy tests

Drop weight testing is preferred to Charpy method for measuring the damage resistance of a
fibre-reinforced polymer matrix composite subjected to an impact event. ASTM D7136 pro-
vides the necessary guidelines for measuring the damage resistance of composites subjected
to drop-weight impact events.
The first attempts to study the impact response of composites focused on the use of the
Charpy impact test method (Cantwell and Morton, 1991). However, it was found that in
Charpy impact testing, unlike for metals and plastics wherein a simple tensile failure could be
induced at the notch root, the composite materials exhibited all three basic failure modes—
tension, compression and shear in the same specimen. Thus, the mode of failure of the com-
posite is inconsistent and it is difficult to relate the measured absorbed energy of the Charpy
test specimen to that of an actual structural component. Thus, after considerable research,
Charpy impact testing of composites was abandoned.
Because a primary concern is potential damage to composite structures induced by low-
velocity impact, as due to a tool drop on an aircraft wing skin or an automobile component
bumping a stationary object, striking a flat plate with a falling weight was subsequently
adopted as the impact test method of choice. Typically, a plate of the order of 150-mm
square is supported around its edges and impacted at its centre. The impacting tup is typi-
cally hemispherical and is in the order of 15 mm diameter. The delivered level of impact
energy is selectable. Therefore, test technicians have the option to ensure full penetration.
Also, the test panel may be subjected to a prescribed level of impact energy, selected to
induce local damage without penetration. When using a drop weight impact tester, two
categories of damage can occur. The first is clearly visible impact damage (CVID), which
can easily be seen by the naked eye. The second type of damage is barely visible impact
damage (BVID), which can seldom be seen by the naked eye. Evaluation of both types
of damage can be done through post-impact compressive and tensile strength tests of
composite specimens subjected to impact. Figures 1 and 2 shows the support fixture for
drop-weight test on a composite laminate and the drop-weight test setup prescribed by
ASTM D7136 respectively.
Figure 1. Support fixture for drop-weight test on composite laminates (Source: ASTM D7136).
61

Figure 2. Impact device with cylindrical tube guide mechanism (Source: ASTM D7136).
3.2 Failure modes of laminated composites under low-velocity impact

Laminated fibre reinforced composite materials have various damage modes such as fibre
breakage, matrix cracking, and delamination. These various damage modes may appear
together under impact loading.
a. Matrix cracking: Matrix cracking is the first type of failure, usually caused by low-velocity
impact and occurs parallel to the fibres due to tension, compression and shearing. The
impact response of the structure is not affected by matrix cracking. It can decrease the
inter-laminar shear and compression strength properties of the resin or the fibre/resin
interface.
b. De-laminations: De-laminations form between the layers in the laminate and is the most
critical damage in low velocity impact events. It may be usually be initiated by matrix
cracks when the threshold energy has been reached or from low energy impact. It can
dramatically reduce the post-impact compressive strength of the laminate. Nikfar and
Njuguna (2014) suggested that de-laminations resulting in composite laminates can be
studied using Compression-after-impact (CAI) tests.
c. Fibre failure: Fibre pull-out and fibre breakage are the most common failures under low
velocity impact testing. Fibre failure occurs because of the high stress field and indenta-
tion effects.
4 FINITE ELEMENT ANALYSIS
In the surveyed literatures, the finite element analysis of the composite laminates were con-
ducted using the commercial code ABAQUS/Explicit. The standard code specifying the
standards and specifications for low velocity impact testing on polymer matrix composites is
ASTM D7136/7136M-12.
4.1 Modelling
Initially, researchers made use of a spring-mass model to analyse the laminate specimens
thereby reducing the degrees of freedom to a finite number making analysis easier. However,
it was found out in the later studies that such an approximation could not account for the
vibration of the specimens resulting during the post-impact stages. Hence, the Hertzian
classical model for contact between an elastic sphere and its half-space was thereafter applied
to study the impact response of composite materials. Oguibe and Webb (1999) noted that
modelling based on the Hertzian contact law is not adequate in describing the contact
behaviour of laminated composite plates on account of its anisotropic nature and high
invariability in properties. The after-math of all these studies caused analysts to resort to
62

numerical techniques as the Finite element method for further studies in relation to impact
on composite laminates.
4.1.1 Plate modelling

Khalili et al. (2001) made use of SC8R elements, an eight-node continuum shell elements,
having 3 degrees of freedom per node for modelling the laminate plates. Oguibe and Webb
(1999) modelled the same using four-node quadrilateral-shell elements. Only one-quarter of
the plate need be modelled due to the symmetry in the laminate geometry, loading, and
boundary conditions.
4.1.2 Impactor modelling

The impactor tup can be modelled via 3 different methods: (1) impactor is assumed as a rigid
body. In this case, no property need be assigned to the impactor in the software which implies that
the impactor possess infinite rigidity, (2) impactor is assumed as a rigid body, but with properties
assigned meaning that the impactor has finite rigidity and (3) it is also possible to consider the
impactor as a deformable body with property assigned. But the 3rd method of impactor model-
ling is highly complicated, making use of C3D8R elements and property assignments as Density
and Young’s modulus value to the impactor. Most of the literatures resorted to assuming the
impactor as a rigid body of infinite rigidity as the deformation of the projectile is negligible and
the analysis using this option is easier and less complicated while giving satisfactory results.
4.1.3 Contact modelling

Khalili et al. (2001) made use of the hard contact law to study the involved processes in the
impactor-laminate contact. In this method, the constraints of contact are applied only when
the separation between the two surfaces, to be in contact, becomes zero. The constraint is,
thereafter, removed when the contact pressure between the two surfaces again becomes zero.
The contact force was measured as a function of the penetration distance. Oguibe and Webb
(1999) made use of the impact algorithm for contact in ABAQUS for simulation of low-
velocity impact. This algorithm is based on the concept that, at the effective point of impact,
the points subjected to contact acquires the same velocity in its direction of impact while
the momentum of both the involved bodies remains unchanged. Modelling of contact also
required the identification of the points or cells of contact in order to evaluate the stresses
and strains at the exact points of contact. For every integration point in each of the 2 con-
tact surfaces, relative displacements of the points are measured to ensure if there has been
a contact or not. If a contact has occurred between the surfaces, Langrangian multipliers
were made use of at each integration point to place the constraint of non-penetration of the
surfaces and its sign was made use of in the detection of surface separation.
4.2 Mesh pattern and convergence study

All literatures made use of finer meshes at and around the vicinity of the contact area in
order to acquire a more accurate quantification of the stresses and strains in this region. The
mesh density was determined by performing repeated analysis with increasing mesh density
until further increase in mesh density resulted in no significant change in results.
5 EXPERIMENTATIONS IN THIS FIELD
Literatures propose conducting experimentation on laminated composite plates in accord-

ance with the specifications laid out in ASTM D7136. The testing principle is based on
Newton’s second law of motion and the principles of conservation of energy. The impact
energy level can be determined by a known mass and the drop height based on energy balance
or it can be identified by measuring the initial velocity of the impactor. Since the impactor
used in this research was of fixed mass, the desired incident impact energy was obtained by
varying the height of the impactor.
63

5.1 Parameters influencing low-velocity impact of composite laminates
A large number of investigations into the parameters that are capable of influencing the dam-
age evolution in composite laminates have been conducted. Various parameters that were
found to have a significant effect on the damage behaviour of composite laminates include:
5.1.1 Influence of impact location

Malhotra and Guild (2014) conducted experiments on laminated composites to determine the
influence of impact location on impact behaviour of glass-fibre reinforced composites. The
impact position employed for the study included central, on-edge and near-edge impacts. The
specimens, after subjecting to impact, were tested for their tensile and compressive strengths
to highlight the different damage mechanisms involved along with a qualitative analysis of
damage propagation by making use of non-destructive evaluation utilizing X-ray tomogra-
phy. It was found that central impact damage was capable of more energy absorption when
compared to on-edge or near-edge impacts since the area available for energy absorption was
more. It was also found that compression after impact tests were capable of highlighting the
de-laminations within the specimens while tension after impact highlights the fibre-breakage
occurring within the specimen.
5.1.2 Influence of temperature

Im et al. (2000) conducted experiments on Carbon-fibre laminated composites utilising a gas
gun with compressed air to subject the laminate specimens to low-velocity impacts at specimen
surface temperatures of -30°C, 20°C, 90°C and 120°C. It was found that as the temperature
increased, the delamination area decreased and vice-versa indicating less damage suscep-
tibility at higher temperatures. Amal Badawy (2012) studied the influence of temperature
using notched Izod impact tests on glass-fibre epoxy specimens at exposure temperatures of
-10°C, 20°C, 50°C and 80°C. He found that fibre fraction volume and temperatures played
an important role in determining the impact response of laminated plates and confirmed less
damage susceptibility of composite laminate specimens at higher temperatures.
5.1.3 Influence of projectile velocity

Mili and Necib (2001) conducted experiments on clamped circular laminated plates to deter-
mine the influence of impactor velocities and stacking sequences on the low-velocity impact
response of the laminated composite plates. It was found that an increase in impact velocity
causes an increase in the amount of impact force as well as the central deflection sustained
by the laminate. However, when the impact velocity decreased, the deflection duration was
found to be more important than the contact time.
5.1.4 Influence of laminate thickness

Amaro et al. (2012) conducted experiments on laminate plates to determine the influence of
thickness on low velocity impact of carbon-fibre epoxy laminates. Plates of different thick-
nesses were tested using an impactor of hemispherical shape. The damage of specimens result-
ing from low-velocity impact mainly constituted of delamination evaluated using ultrasonic
C-scan technique. It was found that the de-laminations increase with increase in plate thickness.
6 CONCLUSION
The main drawback of this area of research is that the laminate failure itself is difficult to
predict due to the highly heterogeneous nature of these materials. Impact causes stiffness
degradation of the laminate specimens. The resulting specimen may or may not have a high
damage tolerance depending on the extent of damage the spontaneous impact might have
caused. This invariability in strength and stiffness after impact damage makes analysis of
composite laminate specimens even more difficult.
Laminated composites may be required to perform their expected functions in environ-
ments where they may be exposed to low energy impact damage from tool dropping, collisions
64

with runway debris, or any similar foreign object impact. Internal damage or flaws in lami-
nated composites may be of microscopic sizes but brings about a high degradation in the
mechanical properties of the laminates making them unsuitable for field applications. These
internal damages may propagate further under repeated low velocity impacts worsening its
suitability for intended use. The properties and behaviour along with techniques of improv-
ing composite structures to improve the impact resilience of laminated composites have been
subjected to study by many researchers. Critical analysis and evaluation of these studies and
techniques involved confirms several constraints in the adopted procedures and remedial
measures as high tool costs, impracticability of studied parameters as temperature which
cannot be maintained at an optimal range during its real-life application thus necessitating
further research in this regard to improve the impact resilience of laminated composites.
REFERENCES
Amal A.M. Badawy. (2012). “Impact behavior of glass fibers reinforced composite laminates at different
temperatures.” Ain Shams Engineering Journal, Vol. 3, pp. 105–111.
Ana M. Amaro, Paulo Nobre Balbis Reis, Marcelo de Moura and Jaime B. Santos. (2012). “Influence
of the specimen thickness on low velocity impact behavior of composites.” Journal of Polymer
Engineering, DOI: 10.1515, pp. 53–58.
ASTM D7136/D7136M-12- “Standard Test Method for Measuring the Damage Resistance of a Fiber-
Reinforced Polymer Matrix Composite to a Drop-Weight Impact Event.”
Belingardi, G., and Vadori, R. (2001). “Low velocity impact tests of laminate glass-fiber-epoxy matrix
composite material plates.” International Journal of Impact Engineering, Vol. 27, pp. 213–229.
Cantwell, W.J., and Morton, J. (1991). “The impact resistance of composite materials—a review.”
Composites, Vol. 22(5), pp. 347–362.
Farooq, U. (2014). “Finite Element Simulation of Flat Nose Low velocity impact behaviour of Carbon
fibre composite laminates.” Unpublished Ph.D. Thesis, School of Engineering and Technology,
University of Bolton, USA.
Khalili, S.M.R., Soroush, M., Davar, A., and Rahmani, O. (2001). “Finite element modeling of low-
velocity impact on laminated composite plates and cylindrical shells.” Composite Structures, Vol. 93,
pp. 1363–1375.
Kwang-Hee Im, Cheon-Seok Cha, Jae-Woung Park, Yong-Hun Cha, and In-Young YangJong-
An Jung. (2000). “Effect of temperatures on impact damage and residual strength of CFRP
composite laminates.” AIP Conference Proceedings, DOI: 10.1063/1.1306184, pp. 1247–1254.
Malhotra, A., and Guild, F.J. (2014). “Impact Damage to Composite Laminates: Effect of Impact
Location.” Appl Compos Mater, Vol. 21, pp. 165–177.
Malhotra, A., Guild, F.J., and Pavier, M.J. (2008). “Edge impact to composite laminates: experiments
and simulations.” Journal of Material Science, Vol. 43, pp. 6661–6667.
Mili, F., and Necib, B. (2001). “Impact behavior of cross-ply laminated composite plates under
low velocities.” Composite Structures, Vol. 51, pp. 237–244.
Nikfar, B., and Njuguna, J. (2014). “Compression-after-impact (CAI) performance of epoxy carbon
fibre-reinforced nanocomposites using nanosilica and rubber particle enhancement.” IOP Conference
Series: Materials Science and Engineering, Vol. 64, DOI: 10.1088/012009.
Oguibe, C.N., and Webb, D.C. (1999). “Finite-element modelling of the impact response of a laminated
composite plate.” Composites Science and Technology, Vol. 59, pp. 1913–1922.
Park, H., Kong, C., Lim, S., and Lee, K. (2011). “A study on impact damage analysis and test of
composite laminates for aircraft repairable design.” 18th International Conference on Composite
materials. August 21–26, 2011, Jeju, Korea.
Pritchard, J.C., and Hogg, P.J. (1990). “The role of impact damage in post-impact compression.”
Composites, Vol. 21(6), pp. 503–511.
Richardson, M.O.W., and Wisheart, M.J. (1996). “Review of low-velocity impact properties of
composite materials.” Composites, Vol. 27 A, pp. 1123–1131.
Shyr, T.W., and Pan, Y.H. (2003). “Impact resistance and damage characteristics of composite
laminates.” Composite Structures, Vol. 62, pp. 193–203.
Zeng, S. (2014). “Characterisation of Low Velocity Impact Response in Composite Laminates.”
Unpublished Ph.D. Thesis, School of Engineering and Technology, University of Hertfordshire, UK.
65

Engineering applications of laminated composites and

various theories used for their response analysis
Raj K. Binia & K.K. Smitha

Department of Civil Engineering, Sree Narayana Gurukulam College of Engineering, Kolenchery,
Ernakulam, India
ABSTRACT: A composite is a material made from two or more constituent materials with
significantly different properties, but when combined, produce a material with characteristics
different from the individual components. These materials are stronger, lighter or economical.
Composites are generally used for buildings, bridges and structures, such as boat hulls,
swimming pool panels, race car bodies, shower stalls, bathtubs, storage tanks, cultured
marble sinks and countertops. Composite laminates are an assembly of layers of fibrous
composite materials which can be joined to provide required engineering properties including
in-plane stiffness, bending stiffness, strength and coefficient of thermal expansion. Layers
of different materials may be used, resulting in a hybrid laminate. The individual layers are
generally orthotropic or transversely isotropic. The theories used for explaining the behavior
of isotropic materials cannot be applied for laminated composites. With the laminate exhibit-
ing anisotropic or quasi-isotropic properties, various other theories were applied in order to
reveal their behavior. In this paper, application of laminated composites in the engineering
field, and various theories used for explaining its behavior, are discussed.
Keywords: laminated composite, nano composite, polymer matrix, metal matrix
1 INTRODUCTION
The idea of combining two or more different materials resulting in a new material with
improved properties has existed for some time. The use of light weight material for different
applications has fascinated mankind. It was discovered that, composite materials have advan-
tages and higher performance compared to each of the individual materials it contain.
Today, composites are being developed that can be cured at low pressures and temperatures,
typically 1 atm and as low as 30°C; An example is the flexible heating jackets suitable for
all plastic, metaland fiber containers. Together with lower cost and easier processing, this
technology is now becoming a practical proposition for consideration by the building industry.
Various two-dimensional plate theories have been developed for modeling laminated
composites and have been broadly classified as an equivalent single layer, layer-wise and
zig-zag theories. The Equivalent Single Layer (ESL) theories are derived by making suitable
assumptions concerning the kinematics of deformation or the stress state through the
thickness of the laminate. The simplest ESL laminated plate theory is the Classical Laminated
Plate Theory (CLPT). The next theory of ESL laminated plate theories is the First-Order
Shear Deformation Theory which extends the kinematics of the CLPT by including a gross
transverse shear deformation in its kinematic assumptions. The First-Order Shear Deforma-
tion Theory requires shear correction factors, which are difficult to determine. Second and
higher-order ESL laminated plate theories use higher-order polynomials in the expansion
of the displacement components through the thickness of the laminate. The higher-order
theories introduce additional unknowns that are often difficult to interpret in physical terms.
67

2 COMPOSITES
Composites are structural materials which consists of two or more combined constituents,
that is, the reinforcing phase and the one in which it is embedded is the matrix phase.
Monolithic metals and their alloys itself cannot always meet the demands of today’s
advanced technologies. By combining several materials, and forming composites, the per-
formance requirements can be met. Composites offer several advantages over conventional
materials, such as improved strength, stiffness, impact resistance, light weight parts, thermal
conductivity, and corrosion resistance. Even though the cost of composite materials is high,
the reduction in the number of parts and its lightweight nature, makes it an economical
option when compared to conventional materials.
A lamina is a thin film of a composite material. Laminating film is offered in a variety of
thicknesses, ranging from thin and flexible to thick and rigid. A laminate is constructed by
stacking a number of such laminae in the direction of the lamina thickness (Figure 1).
3 CLASSIFICATION OF COMPOSITES
Composites are generally classified by their geometry of the reinforcement—particulate,

flake, and fibers and by the type of matrix—polymer, metal, ceramic, and carbon. Further
details are as follows:
Particulate composites consist of particles that are immersed in matrices, such as alloys
and ceramics. It has advantages, such as improved strength, increased operating tempera-
ture, and oxidation resistance.
Flake composites consist of flat reinforcements of matrices. Typical flake materials are
glass, mica, aluminum, and silver. They provide advantages, such as high flexural modulus,
higher strength, and low cost.
Fiber composites consist of matrices reinforced by short (discontinuous) or long
(continuous) fibers. Fibers are generally anisotropic and examples include carbon and
aramids.
Nano-composites consist of materials that are of nanometers (10−9 nm in diameter) in
scale. The accepted range to be classified as a nanocomposite is that one of the constitu-
ents is less than 100 nm length.
The most advanced composites are Polymer Matrix Composites (PMCs) which consist
of a polymer (epoxy, polyester, or urethane, for example) reinforced by thin diameter fibers
(such as graphite, aramids, and boron). They are the most common because of their low cost,
high strength, and simple manufacturing principles.
Metal Matrix Composites (MMCs) have a metal matrix. Examples of matrices include
aluminum, magnesium, and titanium.
Ceramic Matrix Composites (CMCs) have a ceramic matrix, such as alumina calcium and
alumino silicate reinforced by fibers such as carbon or silicon carbide.
Carbon-carbon composites use carbon fibers in a carbon matrix. These composites are
used in very high temperature environments of up to 6000°F (3315°C).
Figure 1. Typical laminate made of three laminae.
68

4 APPLICATION OF COMPOSITES
4.1 Application of nano composites

Nano composite films show enhanced properties, such as elastic modulus, transmission rates for
water vapor and heat distortion. Examples of applications of nano composites include packag-
ing applications for the military, artificial muscles, smart clothes and flexible wires.
4.2 Applications of polymer matrix composites

Their applications range from tennis racquets to the space shuttle. Some of their applications
in various engineering fields are,
Aircraft and space. The percentage of structural weight reduction over metal parts has
been found to be more than 20%. Composites are also used in panels and floorings of
airplanes. Due to high specific modulus and strength, and dimensional stability during
large changes in temperature in space, composites are widely used.
Medical devices. Applications include lightweight face masks, artificial portable lungs and
X-ray tables.
Bridge construction. These are gaining wide acceptance due to their low weight, corrosion
resistance, longer life cycle, and limited earthquake damage.
Automotive. By weight, about 8% of today’s automobile parts are made of composites
including bumpers, body panels and doors.
4.3 Applications of metal matrix composites

Their applications in various engineering fields are:
Space. The space shuttle uses boron/aluminum tubes to support its fuselage frame.
Military. Metal matrix composites such as SiC/aluminum composites satisfy the
dimensional stability requirement because they have high micro yield strength.
Transportation. Metal matrix composites are finding use now in automotive engines that
are lighter than their metal counterparts.
4.4 Applications of ceramic matrix composites

These are finding increased application in high temperature areas in which metal and polymer
matrix composites cannot be used.
4.5 Applications of carbon-carbon composites

Their applications in various engineering fields are:
Space shuttle nose cones. This is because carbon-carbon composites have the lowest overall
weight of all ablative materials, high thermal conductivity to prevent surface cracking, high
specific heat to absorb large heat flux, and high thermal shock resistance to low temperatures
in space. Also, the carbon-carbon nose remains undamaged and can be reused many times.
Aircraft brakes. The carbon-carbon brakes have high durability (two to four times that of
steel), high specific heat (2.5 times that of steel), low braking distances and braking times,
and large weight savings.
5 THEORIES FOR THE ANALYSIS OF LAMINATED COMPOSITES
Analysis of composite material is quite a difficult one. The properties of each layer vary, and as
a whole, the material properties should be analyzed. For this, various theories were formulated.
69

Composite laminates have their planar dimensions one to two orders of magnitude larger
than their thickness. Often laminates are used in applications that require membrane and
bending strengths. Therefore, composite laminates are treated as plate elements.
Various theories were used for the analysis of laminated composites. Among them the
main three were:
Classical Laminated Plate Theory;
First-Order Laminated Plate Theory; and
Higher-Order Laminated Plate Theory.
5.1 Classical laminated plate theory

5.1.1 Assumptions
In the CLPT, it is assumed that the Kirchhoff hypothesis holds:
1. Straight lines are perpendicular to the mid surface (i.e., transverse normals) before defor-
mation and remain straight after deformation.
2. The transverse normals do not experience elongation (i.e., they are inextensible).
3. The transverse normals rotate such that they remain perpendicular to the mid-surface
after deformation.
The first two assumptions imply that the transverse displacement is independent of the
transverse (or thickness) coordinate and the Transverse Normal Strain (εzz) is zero. The third
assumption results in Zero Transverse Shear Strains where εxz = 0 and εyz = 0.
5.1.2 Displacement and strain field

By the Kirchhoff assumptions, a material point occupying the position x, y, z in the unde-
formed laminate, moves to the position x + u, y + v, z + w in the deformed laminate, where
u, v, w are the components of the total displacement. The Kirchhoff hypothesis requires the
displacements u, v, w to be such that:
∂w0
u ( x, y, z, t ) = u0 ( x, y, t ) − z
∂x
∂w0
v ( x, y, z, t ) = v0 ( x, y, t ) − z (1)
∂y
w ( x, y, ) = 0 ( )
, z, t w x, y, t
where u0, v0, w0 are the displacements along the coordinate lines of a material point on the
xy-plane.
For the assumed displacement field in Equation 1, ∂∂wz = 0. The strain equation is:
2
∂u0 1  ∂w0  ∂ 2w0
ε xx = +   − z
∂x 2  ∂x  ∂x 2
1  ∂u ∂v ∂w ∂w  ∂ 2w0
ε xy =  0 + 0 + 0 0  − z
2  ∂y ∂x ∂x ∂y  ∂x ∂y
2
∂v 1  ∂w  ∂ 2w
ε yy = 0 +  0  − z 20 (2)
∂y 2  ∂y  ∂y
1  ∂w ∂w 
ε xz = − 0 + 0 = 0
2  ∂x ∂x 
1  ∂w0 ∂w0 
ε yz = − + =0
2  ∂y ∂y 
ε zz =0
70

5.1.3 Laminate constitutive equations
Although the strains are continuous through the thickness, stresses are not, due to the change
in material coefficients through the thickness (i.e., in each lamina). Hence, the integration of
stresses through the laminate thickness requires lamina-wise integration. The force resultants
are given by:
(0 ) (1)
N xx   A11 A12 A16   ε xx   B11 B12 B16  ε xx 
   
 (0)   
 (1) 
 N yy  =  A12 A22 A26   ε yy  +  B12 B22 B26   ε yy  (3)
N  A A26 
A66  γ xy   B16
( 0 ) B26 
B66  γ xy 
(1)
 xy   16    
(0 ) (1)
M xx   B11 B12 B16   ε xx   D11 D12 D16  ε xx 
   
 (0)   
 (1) 
 M yy  =  B12 B22 B26   ε yy  +  D12 D22 D26   ε yy  (4)
M  B B26  
B66  γ xy   D16
( 0 ) D26  
D66  γ xy 
(1)
 xy   16    
where Aij is the Extensional Stiffness, Dij the Bending Stiffness, and Bij the Bending-Extensional
Coupling Stiffness, which are defined in terms of the Lamina Stiffness ( Qij(k ) ) as:
h
2 N zk +1
( ) ( )
Aij , Bij , Dij = ∫ Qij 1, z, z 2 dz = ∑ ∫ Qij(k ) 1, z, z 2 dz
k =1 zk
( )
−h
2
N
Aij = ∑Qij(k ) ( zk +1 − zk ) , (5)

k =1
1 N
(
Bij = ∑Qij(k ) zk2 +1 − zk2
2 k =1
)
1 N
(
Dij = ∑Qij(k ) zk3 +1 − zk3
3 k =1
)
Note that Qij, and therefore Aij, Bij and Dij, are, in general, functions of position (x, y).
Equations can be written in a compact form as:
 N  [ A] [B ] ε 0 

  =  (6)
M  [ B ] [D ] ε 1 
where ε 0 and ε 1 are defined as Vectors of the Membrane and Bending strains, respectively.
5.2 First-order laminated plate theory

The CLPT based on Kirchhoff’s hypothesis is less accurate as it neglects the transverse shear
effects. The hypothesis states that the transverse normal to the midplane of the plate remains
straight and normal after deformation because of the negligible transverse shear effects. To
overcome this limitations of CLPT, the First-Order Shear Deformation Theory (FSDT) was
formulated.
5.2.1 Displacements and strains field

In the FSDT, the Kirchhoff hypothesis is relaxed by removing the third part, that is, the trans-
verse normals do not remain perpendicular to the mid-surface after deformation. Transverse
shear strains are included in the theory. The inextensibility of transverse normals requires
that w not be a function of the Thickness Coordinate, z. Under the same assumptions and
restrictions as in the CLPT, the displacement field of the FSDT is of the form:
71

u ( x, y, z,t ) = u0 ( x, y,t ) + zØ x ( x, y,t )
v ( x, y, z,t ) = v0 ( x, y,t ) + zØ y ( x, y,t ) (7)
w ( x, y, z,t ) = w0 ( x, y,t )
where u0, v0, w0, Øx, Øy are unknown functions to be determined, the nonlinear strains associ-
ated with the displacement field are:
2
∂u0 1  ∂w0  ∂Ø x
ε xx = +   +z
∂x 2  ∂x  ∂x
 ∂u0 ∂v0 ∂w0 ∂w0   ∂Ø x ∂Ø y 
γ xy =  + +  + z +
 ∂y ∂x ∂x ∂y   ∂y ∂x 
2
∂v 1  ∂w  ∂Ø y
ε yy = 0 +  0  + z (8)
∂y 2  ∂y  ∂y
∂w0 ∂w
γ xz = + Ø x , γ yz = 0 + Ø y ,
∂x ∂y
ε zz = 0
The Strains (εxx, εyy, γxy) are linear through the laminate thickness, while the Transverse
Shear Strains (γxy, γyz) are constant through the thickness of the laminate in the FSDT.
5.2.2 Laminate constitutive equations

When thermal and piezoelectric effects are not present, the Stress Resultants (N’s and M’s)
are related to the Generalized Displacements (u0, v0, w0, Øx, Øy) by the relations:
 ∂u0 1  ∂w0  2   ∂Ø x 
 +     
 ∂x 2  ∂x   B ∂x
N xx   A11 A12 A16    11 B12 B16   
   
2
     ∂v0 1  ∂w0   ∂Ø y
 N yy  =  A12 A22 A26   +  +
  B12 B22 B26    (9)
N  A  ∂y 2  ∂y   B  ∂y 
 xy   16 A26 A66  B26 B66 
 ∂u ∂v ∂w ∂w   16  ∂Ø ∂ Ø 
 0
+ 0
+ 0 0
  x+
y

 ∂y ∂x ∂x ∂y   ∂y ∂x 
 ∂u0 1  ∂w0  2   ∂Ø x 
 +     
 ∂x 2  ∂x   D ∂x
M xx   B11 B12 B16     11 D12 D16   
 
2
     ∂v0 1  ∂w0    ∂Ø y
 M yy  =  B12 B22 B26   +    +  D12 D22 D26    (10)
M  B ∂ y 2  ∂ y ∂ y
 xy   16 B26 B66    D D26 D66   
 ∂u ∂v ∂w ∂w   16  ∂Ø ∂ Ø 
 0+ 0+ 0
y
0
  x+ 
 ∂y ∂x ∂x ∂y   ∂y ∂x 

 ∂w0 
+ Øy 
Qy   A44 A45   ∂y 
 =K   (11)
Qx   A45 A55   ∂w0 
+ Øx
 ∂x 
5.3 Higher order laminated plate theory

The FSDTs assume linear in-plane stresses and displacements, respectively, through the
laminate thickness. Since FSDT accounts for layer-wise constant states of transverse shear
stress, FSDT requires a shear correction factor in order to satisfy the traction free boundary
conditions at top and bottom surfaces of the plate. To overcome the limitation of FSDT,
72

higher-order theories were developed. Higher-Order theories can represent the kinematics
better, may not require shear correction factors, and can yield more accurate inter-laminar
stress distributions. However, they involve higher-order stress resultants that are difficult to
interpret physically and require considerably more computational effort.
In higher-order theories, an additional dependent unknown is introduced into the theory
with each additional power of the thickness coordinate. In addition, these shear deformation
theories do not satisfy the conditions of zero transverse shear stresses on the top and bottom
surfaces of the plate, and require a shear correction to the transverse shear stiffness.
5.3.1 Third-order plate theory: Displacement field

The Third-order Plate Theory developed was based on the same assumptions as the classical
and first-order plate theories, except that the assumption on the straightness and normality
of a transverse normal after deformation by expanding the Displacements (u, v, w) as cubic
functions of the thickness coordinate was relaxed.
The displacement field can be expressed in terms of u0, v0, w0, Øx, Øy:
4 3 ∂w 
u ( x, y, z,t ) = u0 ( x, y,t ) + zØ x ( x, y,t ) − z  Øx + 0 
3h 2  ∂x 
4 3 ∂w0 
v ( x, y, z,t ) = v0 ( x, y,t ) + zØ y ( x, y,t ) − 2 z  Ø y + (12)
3h  ∂y 
w ( x, y, z,t ) = w0 ( x, y,t )
5.3.2 Strains and stresses field

Substitution of the displacement equations into the nonlinear strain displacement relations
yields the strains:
(0 ) ε xx(1)  (3) 
ε xx   ε xx   ε xx

   (0 )    (1)  3  (3) 
  
 ε yy  =  ε yy  + z  ε yy  + z  ε yy 
γ   (0)   (1)   (3) 
 xy  γ xy  γ xy  γ xy  (13)
( 0) ( 2)
γ yz  γ yz  2 γ yz 
  =  (0 )  + z  (2 ) 
γ xz  γ xz  γ xz 
 4 
where  c1 = 3c1 and c1 = 2 
 3h 
 ∂uo 1  ∂wo   2
 +   
 ∂x 2  ∂x  
( 0)
 ε xx   2 
 ( 0)   ∂vo 1  ∂w
wo  
 ε yy  =  +    (14)
 γ ( 0)   ∂y 2  ∂y  
 xy   ∂uo ∂vo ∂wo ∂wo 
 + + 
 ∂y ∂x ∂x ∂y 

 ∂∅ x 
 
( 1)  
 ε xx ∂x   ∂wo 
( 0)   ∅ y +
 ( 1)   ∂∅ y
  γ
 yz   ∂y 
 ε yy  =   , ( 0 )  =   (15)
 γ ( 1)   ∂y   γ xz   ∅ + ∂wo 
 xy   ∂∅ ∂∅ y  
x
∂x 
 x
+ 
 ∂y ∂x 
73

 ∂∅ x ∂ 2wo 
 + 
( 3) 
 ε xx  ∂x ∂x 2
  ∂wo 
 ( 3)   ∂∅ y ∂ 2wo   γ (yz2 )   ∅ y + ∂y 
 ε yy  = − c1  + ,  ( 2 )  = − c2   (16)
 γ ( 3)   ∂y ∂y 2
  γ xz   ∅ + ∂wo 
 xy   ∂∅ ∂ 2wo   ∂x 
x
∂∅ y
 x
+ +2 
 ∂y ∂x ∂x∂y 
To validate these various theories, a model of pressure was examined analytically and
numerically by the use of Ansys16.0 software:
5.3.2.1 Analytical solution
Here, D was the Mean Diameter of a vessel (450 mm), t was Thickness of the vessel (10 mm),
and the length of the vessel was 4 m. The Pressure was P = 10 kN.
PD 10 × 450
Hoop stress: σ 0 = = = 225 MPa
2t 2 × 10
5.3.2.2 Numerical solution

Element type, 8 NODE 281, was used based on the First-Order Shear Deformation Theory.
Load was uniformly distributed along the internal area. Material Property for the steel was:
E = 200000 Mpa µ = 0.3
Figures 2 and 3 shows the deformed and stress at x direction. Taking values at the central
nodes of the cylinder, the hoop stress values obtained were:
σ x = 3.9123 Mpa and σ y = 216.1725 MPa
Hoop stress = σ x2 + σ y2 = 216.20 MPa
Figure 2. Deformed shape of pressure vessel. Figure 3. Stress along x direction.
Table 1. Stress value comparison.
Analytical From %
Stress solution software variation
Hoop stress 225 MPa 216.20 MPa 3.8%
74

Table 1 shows the percentage variation of stress values using analytical and numerical solu-
tion (Software).
6 CONCLUSIONS
Composite materials are using widely today because of their improved strength, stiffness,
reduced weight, reduced life cycle cost and so on. Composites are now extensively being used
for rehabilitation/strengthening of preexisting structures that have to be retrofitted to make
them seismic resistant, or to repair damage caused by seismic activity.
Composites form a new material with improved properties above that of the individual
materials themselves. The fabrication of composite and its analysis needed a careful study
to explain its structural properties. Various theories were used in the analysis of composite
materials; among them Equivalent Single Layer (ESL) theories were used to explain the plate
composites.
The ESL theory includes the Classical Lamination Plate Theory (CLPT), the First-Order
Shear Deformation Theory (FSDT), and the Higher-Order Shear Deformation Theory
(HSDT).
CLPT is less accurate as it neglects the transverse shear effects. To overcome these limita-
tions, FSDT was proposed. It requires the use of a shear correction factor in order to satisfy
the traction free boundary conditions at top and bottom surfaces of the plate. The accuracy
of the response by FSDT strongly depends upon the choice of shear correction factors. To
overcome the limitations of FSDT, HSDT involving a transverse shear stress function, was
developed. The HSDT introduced additional unknowns.
By examining numerically a model for a pressure vessel using Ansys APDL 16.0, the theo-
ries were validated. This problem was based on a First-Order Shear Deformation Theory.
REFERENCES
Auricchio, F. & Sacco, E. (2003). Refined first-order shear deformation theory models for composite
laminates. Journal of Applied Mechanics, 70, 381.
Jauhari, N., Mishra, R. & Thakur, H. (2017). Failure analysis of fibre-reinforced composite laminates.
Materials Today: Proceedings, 4, 2851–2860.
Kant, T. & Swaminathan, K. (2002). Analytical solutions for the static analysis of laminated composite
and sandwich plates based on a higher order refined theory. Composite Structures, 56, 329–344.
Kaw, A.K. (2006). Mechanics of Composite Materials (2nd ed.).: (London, New York), Taylor &
Francis.
Kharghani, N. & Guedes Soares, C. (2016). Behaviour of composite laminates with embedded
delaminations. Composite Structures 150, 226–239.
Kumar, A. & Chakrabarti, A. (2017). Failure analysis of laminated skew laminates. Procedia Engineer-
ing, 173, 1560–1566.
Rastgaar Aagaah, M., Mahinfalah M., Mahmoudian, N. & Nakhaie Jazar, G. (2002). Modal analysis of
laminated composite plates using third order shear deformation theory.
Reddy, J.N. (1984). A simple higher-order theory for laminated composite plates. Journal of Applied
Mechanics, 51, 745.
Reddy, J.N. (2004). Mechanics of laminated composite plates and shells theory and analysis (2nd ed.).
(London, New York): CRC Press LLC.
Shokrieh, M.M., Akbari, S. & Daneshvar, A. (2013). A comparison between the slitting method and
the classical lamination theory in determination of macro-residual stresses in laminated composites.
Composite Structures, 96, 708–715.
Singh, D.B. & Singh, B.N. (2017). New higher order shear deformation theories for free vibration and
buckling analysis of laminated and braided composite plates. International Journal of Mechanical
Sciences 131, 265–277.
75

Hydraulics and water resources engineering

Trend analysis of rainfall projections of Teesta river basin,

Sikkim using non-parametric tests and ensemble empirical
mode decomposition
S. Soorya, S. Adarsh & K.L. Priya

TKM College of Engineering, Kollam, Kerala, India
ABSTRACT: This paper presents the spatio-temporal trends and dominant change points
in the Regional Climate Model system (RegCM) simulations at 0.43° × 0.43° resolution over
Teesta river basin located in Sikkim. In this study, first the bias correction is done following
the Local Intensity Scaling (LOCI) method by comparing the interpolated model values with
the observed values of five stations for the historical period of 1983–2005. The estimated
correction is applied to the interpolated future RegCM data for the rainfall projections
of 2021–2050 period under two Representative Concentration Pathway (RCP) scenarios
(RCP4.5 and RCP8.5). The historical rainfall and rainfall projections of the near future period
of 2021–2050 are subjected to trend analysis using Mann-Kendall method and its sequential
version (SQMK), Sen’s slope estimator, linear trend fitting and Ensemble Empirical Mode
Decomposition methods. The preliminary estimates using non-parametric tests showed a
likely reversal in nature of trend in Gangtok and Kalimpong stations. The results of SQMK
test detected possible trend turning points in 2030s except in future rainfall of Darjeeling
station. The results of non-linear trend analysis showed that the future rainfall of Jalpaiguri
shows an increasing trend while that of Gangtok shows decreasing trend irrespective of the
two candidate RCP scenarios. The non-linear trend is different from linear trend in rainfall of
Lachung and Kalimpong stations under both scenarios and that of Jaipalguri station under
RCP8.5 scenario. The extraction of true shape of inherent non linear trend performed in this
study may help for improved predictability and better management of water resources of
Teesta river basin.
1 INTRODUCTION
Trend analysis of rainfall is a major concern among the hydrologists because of the recent
evidences of global climate change. Performing such analysis for projected climate scenarios
of a river basin or geographical region of interest can help in better preparedness against
natural disasters like droughts. A large number of studies were conducted to detect the trends
of hydro-climatic variables in different parts of India at subdivisional scale and river basin
scale (Guhatakurta and Rajeevan 2007; Kumar et al., 2010; Jain and Kumar 2013; Jain et al.,
2013; Adarsh and Janga Reddy 2015). The north eastern part of India and eastern Hima-
layan region are vulnerable to climatic change effects and scientific community is consistently
monitoring such changes. For example, Jain et al. (2013) performed the trend analysis of
for the historical records of 1871–2008 period of rainfall in different subdivisions in north
eastern India. Singh and Goyal (2016a,b, 2017) focused on the climate change projections
of eastern Himalayan regions and upper Teesta watershed using different datasets and tech-
niques. However, they never attempted a formal trend analysis of projected rainfalls.
79

A formal trend analysis of projected rainfall at river basin scale may eventually help
for preparedness and management of water resources. For analyzing the trend of time
series different methods are available including non parametric tests, fittings and time or
frequency domain approaches (Sonali and Nagesh Kumar, 2013; Sang et al., 2013). The
non-parametric tests Mann-Kendall (MK) test (Mann 1945; Kendall 1975) and Sen’s slope
(SS) estimator (Sen 1968) are most popular among them and the sequential version of MK
(SQMK) test may also help to identify the trend turning points in the time series (Chat-
terjee et al., 2012). Due to the changing climate and the induced non-stationarity, it is
important to extract the inherent non-linear trend in the datasets (Franske et al., 2014). For
the extraction of inherent non linear trend, the techniques like singular spectrum analysis,
empirical mode decomposition (EMD) or its variants, wavelet transform etc. have been
used by the researchers (Wu et al., 2007; Sang et al., 2014; Carmona and Poveda 2014;
Unnikrishnan and Jothiprakash 2016; Sang et al., 2017). Moreover, even though Mann-
Kendall and Sen’s slope estimators are still the most popular trend detection methods,
it is reported that the length of the dataset can significantly influence the trend and the
decomposition based methods possess the advantage that it can portray the true shape of
the inherent trend of the dataset (Sang et al., 2014). Owing to the insensitiveness of con-
trol parameters and data adaptive characteristics, this study uses a noise assisted variant
of EMD namely Ensemble EMD (EEMD) for extracting the non-linear trend of histori-
cal and projected rainfall datasets of Teesta basin. In the following section, the details of
EMD are presented briefly. Thereafter, the description of dataset used is presented. Sec-
tion 4 presents the results and discussion. In the final section, important conclusions drawn
from the study are compiled.
2 METHODOLOGY
2.1 Local Intensity Scaling (LOCI)

As there exists inherent bias between Regional Climate Model (RCM) simulations and
observed rainfall, it is to be corrected with respect to the observed data of historical period
in order to get the rainfall projections at gauge locations. Many methods like delta change
approach, analog methods, quantile mapping local intensity scaling (LOCI) method etc.
are available for correcting the bias and in this study, LOCI proposed by Widmann and
Bretherton (2000) method adjust the mean as well as both wet-day frequencies and wet-day
intensities of precipitation time series separately in three steps:
1. First, an RCM-specific control precipitation threshold (Pth,contr) is calibrated such that the
number of RCM-simulated days exceeding this threshold matches the number of observed
days with precipitation larger than 0 mm. Then, the number of precipitation events for
both control and scenario run are corrected by applying the calibrated RCM precipitation
threshold (Pth,contr) so that all days with precipitation less than Pth,contr are redefined to dry
days with 0 mm precipitation:
0, if Pcontr (d ) < Pth,contr 

*1
Pcontr (d ) =   (1)
Pcontr (d ), otherwise 
0, if Pscen (d ) < Pth,contr 
*1
Pscen (d ) =   (2)
Pscen (d ), otherwise 
This procedure allows the scenario run to have a different wet-day frequency than the
control run. In our study, the amount of precipitation on these days was not redistributed
to the remaining rainy days.
2. In a second step, a linear scaling factor is estimated based on the long-term monthly
mean wet-day intensities. Taking only wet days into account (i.e., the observed days with
80

precipitation larger than 0 mm) and the RCM-simulated days with precipitation larger
than the adjusted precipitation threshold (Pth,contr), the intensity scaling factor (s) is calcu-
lated by the following ratio:
µm ( Pobs (d ) / Pobs (d ) > 0 mm )

s= (3)
µm ( Pcontr (d ) / Pcontr (d ) > Pth,contr ) − Pth,contr
*
Pcontr (d ) = Pcontr
*1
(d ).s (4)
where µm is the mean within monthly interval.

3. Finally, the RCM-simulated precipitation is corrected as follows:
*
Pscen (d ) = Pscen
*1
(d ).s (5)
By definition, the adjusted control and scenario precipitation both have the same mean,
wet-day frequency and intensity as the observed time series.
2.2 Ensemble Empirical Mode Decomposition (EEMD)

Empirical Mode Decomposition (EMD) is a signal decomposition method proposed by
Huang et al., (1998), which decomposes a time series signal into different oscillatory modes
having specific periodicity. The different steps involved in the process are:
(1) Identify all extrema (maxima and minima) of the signal X(t); (2) Connect these
maxima with an appropriate interpolation function (for example, cubic spline) to con-
struct an upper envelope (Emax(t)) and lower envelope ((Emin(t)) by connecting maxima
points and minima points respectively; use the same procedure for minima to construct a
lower envelope (3) Compute the mean of the upper and lower envelope, m(t); (4) Calculate
the difference time series d(t) = X(t) - m(t); (5) repeat steps (1) to (4) until d(t) becomes a
zero-mean series with no riding waves (i.e., there are no negative local maxima and positive
local minima) with smoothened amplitudes. Such an oscillatory signal is called an Intrinsic
Mode Function (IMF). This iterative procedure of extraction of oscillatory modes (IMFs
or residue) from a signal is known as ‘sifting’. To satisfy step (5), an appropriate criterion
is to be applied to stop the sifting iterations and the popular modified Cauchy type stop-
ping criterion (Huang and Wu 2008) in which the squared difference between signals of
adjacent iterations is less than a specified tolerance limit (say 0.2–0.3). Then the resulting
signal can be designated as the first intrinsic mode function IMF1 (6) Compute the residue
R1(t) by subtracting IMF1 from original signal (i.e., R1(t) = X(t) - IMF1(t) is used as new
signal. The ‘sifting’ process is repeated upon this signal to get IMF2. (7) To get the higher
order oscillatory modes, this process is repeated upon the ‘difference’ signal obtained by
subtracting all modes from the original signal. The process will be continued till the result-
ing residue is a monotonic function or a function having only one extrema. During the
decomposition of a signal into IMFs, sometimes it may fail to assign the signals with dis-
tinctly different frequencies into separate IMFs. As a result, each IMF contains different
modes of oscillations, which make IMFs to lose physical meaning or falsely represent the
physical processes associated with mode. To solve this problem, a noise assisted data analy-
sis method was proposed by different researchers in the past (Wu and Huang 2005; Torres
et al., 2011). The Ensemble EMD (EEMD) propounded by Wu and Huang (2005) falls in
this category. In the EEMD, the EMD operation is performed upon an ensemble realiza-
tions of size M which are formulated by adding white noise series to the signal such that
X m (t ) = X (t ) + wm (t ), where Xm(t) indicates the mth ‘artificial’ signal; X(t) is the true signal;
and wm (t ) represents the zero mean unit variance white noise signal for the mth realization,
m = 1, 2,…M. Finally, the ensemble mean of the corresponding IMFs is considered as
final IMF.
81

Figure 1. Location map of Teesta basin with CORDEX grids overlapped.
3 STUDY AREA AND DATA
Teesta river basin (latitude of 26°42’N - 28°06’N, and longitude of 88°03’E - 88°58’E), a cli-
matically sensitive basin located within the Eastern Himalayas, Sikkim, India is selected for
the present study. Major portion of the basin is hilly region having an elevation ranging from
285 m to 8586 m. It is bounded by Himalayas in north, east and west. Out of the total area
(11,650 km2) nearly 60% of the catchment area occupies in Sikkim, while rest of the parts
falls in Darjeeling district of West Bengal and northern part of Bangladesh. The observed
rainfall data of stations Gangtok, Jalpalguri, Kalimpong, Lachung and Darjeeling for 1983–
2005 are collected from India Meteorological Department (IMD). It is noted that the north
western part of the catchment is mostly fed by snow and glaciers (Singh and Goyal 2017)
and raingauge installations are scarce in this part of the basin which is unique character of
the basin. The daily rainfall data for three different RCMs (HadGEM3-RA, YSU-RSM,
RegCM) for historical period and for projection 2021–2050 are collected from Coordinated
Regional Climate Downscaling Experiment (CORDEX) database (http://www.cordex.org/).
4 RESULTS AND DISCUSSIONS
In this study, first the grid point values of CORDEX database are interpolated to the gauge
points using the inverse distance weighted (IDW) method as the grid points of former data-
base is not coinciding with that of latter. The bias between RCM simulations and observed
rainfall are corrected by LOCI method.
The monthly averaged values of corrected mean precipitation of the basin for historical
period (1983–2005) are first checked for quality assessment by comparing the monthly mean
rainfall of three candidate RCM simulations with observed data. It was assessed that among
different RCMs the interpolated values of RegCM is showing the best similar pattern of
actual observed rainfall. So in this study, only RegCM data is considered for subsequent
analysis. The LOCI corrections applied for different monthly values are given in Figure 2.
It is noted that the corrections for rainfall are multiplying factors, which are observed to be
positive for all the five stations. The largest correction is obtained for Lachung station (for
the month of May), whereas the smallest correction is obtained for Jaipalguri (for the months
of December and February).
The interpolated future monthly rainfall values are estimated for all the stations for
2021–2050 period, applying the respective corrections for two representative concentration
pathways (RCP) scenarios RCP4.5 and RCP8.5.
82

Figure 2. LOCI corrections (scaling factors) of monthly rainfall of different stations of Teesta basin.
Table 1. Results of MK test and SS estimator of trend analysis of historical and projected rainfall of
different stations in Teesta basin.
Historical (1983–2005) RCP4.5 (2021–2050) RcP8.5 (2021–2050)

Station MK value SS MK value SS MK value SS
Lachung -1.479 -26.14 -0.85 -1.88 -0.12 -0.68

Kalimpong 1.849 28.74 -0.21 -1.38 -0.25 -1.36
Darjeeling 1.268 27.31 0.07 1.02 0.21 1.61
Jaipalguri 0.159 2.40 0.50 4.80 0.25 2.48
Gangtok 1.479 16.65 -0.82 -1.91 -0.68 -6.98
In the trend analysis, the MK test and Sen’s slope tests of the historical datasets (1983–
2005), rainfall projections under RCP4.5 and RCP8.5 scenario of all the five stations are
performed at 5% significance level. The results are presented in Table 1.
From Table 1 it is noted that the trend of historical rainfall records and projections under
both scenarios of Lachung station are alike and shows a decreasing trend. Also, in the his-
torical period, the rainfall of Kalimpong and Gangtok was showing an increasing trend while
there is a likely reduction of rainfall in these stations in the near future (2021–2050). But it is
to be noted that none of the trend is significant at 5% level.
To capture the variability of trend in different decades of near future period and for locat-
ing the trend turning points, an indepth sequential trend analysis is made using SQMK test.
The details of SQMK test can be found elsewhere (Sneyers 1990; Adarsh and Janga Reddy
2015). The results of SQMK test of rainfalls under RCP4.5 are provided in Figure 3 and that
of rainfalls under RCP8.5 sceanario are provided in Figure 4.
The results of SQMK test showed that the rainfall trend of none of the stations is not
significant under RCP4.5 and RCP8.5 scenarios in near future except for some localized time
spells under RCP8.5 scenarios for the data of Kalimpong and Gangtok during 2040–45. But
the rainfall trend in all of the stations shows a possible reduction since 2030s. The trend of
rainfall under both of the scenarios of Darjeeling shows an erratic pattern and there is no
definite trend turning point could be identified for the rainfall data of this station. The possi-
ble trend turning points of Kalimpong and Gangtok occurs nearly between 2030 and 2035.
Further the linear trend fitting of different rainfall series is made and the non-linear trend
is extracted for by invoking the EEMD algorithm. In the EEMD implementation a noise
standard deviation of 0.2 and ensemble size of 300 is used. In all cases 4 or 5 modes evolved
which represents the rainfall variability, which is less than the maximum number expected
log(N), where N is the data length (23 for historical and 30 for projected rainfall in this study).
83

Figure 3. Results of SQMK test of rainfall projections of different data for RCP4.5 scenario. The ver-
tical bar shows a trend turning point. u(t) refers the progressive series and u’(t) refer the retrograde series.
Figure 4. Results of SQMK test of rainfall projections of different data for RCP8.5 scenario. The verti-
cal bar show a trend turning point. u(t) refers the progressive series and u’(t) refer the retrograde series.
Figure 5. Linear and non-linear trend of rainfall projections of Teesta basin for 2021–2050 period
under RCP4.5 scenario. The upper panels of each case show the data along with linear trend fitting. The
lower panel shows the corresponding non-linear trend.
Also only one component evolved with the characteristics of the ‘residue’ and the rest of
them possesses the properties of an IMF (see Huang et al., 1998). Therefore it is believed that
the decomposition is reliable and not a case of ‘over decomposition’.
The results of liner trend and EEMD based non-linear trend of rainfall projections under
RCP4.5 and RCP8.5 are presented in Figures 5 and 6 respectively.
Figure 5 show that the linear trend of rainfall projections of Kalimpong, Darjileeng and
Jaipalguri stations display an increasing trend while for the rest of the stations the trend is
decreasing. On examining the corresponding non-linear trend, it is noticed that the non-
linear trend matches with the linear trend except for Lachung and Kalimpong stations.
Rainfall trend of Kalimpong station shows decreasing trend till 2034 thereafter there is a
likely increase. Figure 6 show that the linear trend of rainfall projections of all the stations
displays a decreasing character but the true non-linear trend of rainfall projections of
different stations show a distinctly different character. The trend of Darjeeling and Gangtok
is monotonically decreasing, while that of Jaipalguri is increasing. The rainfall trend of
84

Figure 6. Linear and non-linear trend of rainfall projections of Teesta basin for 2021–2050 period
under RCP8.5 scenario. The upper panels of each case show the rainfall along with linear trend fitting.
The lower panel shows the corresponding non-linear trend.
Lachung and Kalimpong shows a reduction upto 2030s and thereafter it is increasing. There
is a high level of geographical heterogeneity exists in the region and which is reflected in the
precipitation trends.
The trend analysis showed that there are distinct differences in the projected precipitation
pattern of Teesta basin. There exists large differences in elevation within the basin, with high
altitudinal zones (elevation of ∼ 7000 m) in the upstream portion and low altitude zones
(elevation as low as 1400 m) in the lower catchment. Hence within the study area, the major
driving factor behind the precipitation (such as orography or convection) differs significantly,
which reflected in the scenario based precipitation projections.
The EEMD method is superior in trend analysis of hydro meteorological variables, as it
provides information of non-linear trend, which may improve the predictability efforts. The
present study gives broad inferences of changing climate of Teesta basin under two RCP
scenarios. The study provides a background for hydrologic data generation, detailed inves-
tigations on climate change, framing adaptation policies and hence it may help for overall
management of water resources of Teesta basin in north east India.
5 CONCLUSIONS
This paper presented the trend analysis of rainfall projections of Teesta basin in north-east
India under RCP4.5 and RCP8.5 scenarios using the MK test, Sen’s slope estimator, sequen-
tial MK test, linear fitting and EEMD methods. The important conclusions framed from the
study are:
The non-parametric tests detected an increasing trend in the past and there is a likely
reduction in future rainfall of Kalimpong and Gangtok stations, while the nature of trend
of rest of the stations remains un altered
There is no possible trend turning point in Darjeeling station in either of the RCP sce-
narios in near future during 2021–2050
The trend turning points are noticed in similar years (2030s) in the rainfall of Kalimpong
and Gangtok stations under both RCP scenarios
There is a high level of geographical heterogeneity exists in the study area and which inturn
influences the major driver behind precipitation. It resulted in the differences in the trends
in projected precipitation of the basin
The non-linear trend is different from linear trend in rainfall of Lachung and Kalimpong
stations under both of the RCP scenarios and that of Jaipalguri station under the
RCP8.5 scenario
The true trend of rainfall projections of Kalimpong station under both RCPs and that
in rainfall projections of Lachung station under RCP8.5 scenario displayed distinct non-
linearity. The EEMD method captured such trends successfully and it may eventually help
in accurate simulation or forecasting of rainfall of these stations.
85

REFERENCES
Adarsh, S. & Janga Reddy, M. 2015. Trend analysis of rainfall in four meteorological subdivisions in
Southern India using non parametric methods and discrete wavelet transforms. International Journal
of Climatology 35.6: 1107–1124.
Carmona, A.M. & Poveda, G. 2014. Detection of long-term trends in monthly hydro-climatic series of
Colombia through Empirical Mode Decomposition. Climatic Change 123.4: 301–313.
Chatterjee S., Bisai D. & Khan A. 2012. Detection of approximate potential trend Turning Points in
temperature time series (1941–2010) for Asansol weather observation station, West Bengal, India.
Atmospheric and Climate Sciences 4: 64–69.
Franske, C.L. 2014. Non-linear climate change. Nature Climate change 4: 423–424.
Guhathakurta, P. & Rajeevan, M. 2007. Trends in the rainfall pattern over India. International Journal
of Climatology, 28.11: 1453–1469.
Huang, N.E., Shen, Z., Long, S.R., Wu, M.C., Shih, H.H., Zheng, Q., Yen, N.C., Tung, C.C., & Liu,
H.H. 1998. The empirical mode decomposition and the Hilbert spectrum for nonlinear and non-
stationary time series analysis. Proceedings of Royal Society London, Series A. 454: 903–995.
Huang, N.E., & Wu, Z. 2008. A review on Hilbert Huang Transform: Method and its applications to
geophysical studies. Reviews of Geophysics 46(2), doi: 10.1029/2007RG000228.
Jain, S.K. & Kumar, V. 2012. Trend analysis of rainfall and temperature data for India-A Review.
Current Science, 102.1: 37–49.
Jain, S.K., Kumar V. & Saharia, M. 2013. Analysis of rainfall and temperature trends in north-east
India”, International Journal of Climatology 33.4: 968–978.
Kendall, M.G. 1975. Rank Correlation Methods, 4th Edition, Charles Griffin, London, UK, 1975.
Kumar, V., Jain S.K. & Singh, Y. 2010. Analysis of long-term rainfall trends in India. Hydrological
Sciences Journal 55.4: 484–496.
Mann, H.B. 1945. Non-parametric tests against trend. Econometrica, 13.3: 245–259.
Sang, Y.F., Wang Z., & Liu, C. 2014. Comparison of the MK test and EMD method for trend
identification in hydrological time series. Journal of Hydrology, 510: 293–298.
Sang, Y.-F., Wang Z. & Liu C. 2013. Discrete wavelet-based trend identification in hydrologic time
series. Hydrological Processes 27(14): 2021–31.
Sang Y.-F., Sun F., Singh V.P., Xie P. & Sun J. 2017. A discrete wavelet spectrum approach to identifying
non-monotonic trend pattern of hydro-climate data. Hydrology and Earth System Science Discussions
doi:10.5194/hess-2017-6.
Sen, P.K. 1968. Estimates of the regression co-efficient based on Kendall’s tau. Journal of the American
Statistical Association 63: 1379–1389.
Singh, V. & Goyal, M.K. 2016a. Changes in climate extremes by the use of CMIP5 coupled climate
models over eastern Himalayas. Environmental Earth Science 75: 839.
Singh, V. & Goyal, M.K. 2016b. Analysis and trends of precipitation lapse rate and extreme indices
over north Sikkim eastern Himalayas under CMIP5ESM-2M RCPs experiments. Atmospheric
Research 167 (2016): 34–60.
Singh, V. & Goyal, M.K. 2017. Spatio-temporal heterogeneity and changes in extreme precipitation over
eastern Himalayan catchments India. Stochastic Environmental Research and Risk Assessment. DOI
10.1007/s00477-016-1350-3.
Sneyers, R. 1990. On the statistical analysis of series of observations. Tech. Note 143, 192 pp., Geneva,
Switzerland.
Sonali, P. & Nagesh Kumar, D. 2013. Review of trend detection methods and their application to detect
temperature change in India. Journal of Hydrology 476: 212–227.
Torres, M.E., Colominas, M.A., Schlotthauer, G. & Flandrin, P. 2011. A complete ensemble empirical
mode decomposition with adaptive noise. IEEE International conference on Acoustic Speech and
Signal Processing, Prague 22–27 May 2011, pp. 4144–4147.
Unnikrishnan, P. & Jothiprakash, V. 2015. Extraction of nonlinear trends using singular spectrum
analysis. Journal of Hydrologic Engineering 10.1061/(ASCE)HE.1943-5584.0001237, 05015007.
Widmann, M. & Bretherton, C.S. 2000. Validation of mesoscale precipitation in the NCEP reanalysis
using a new grid cell dataset for the Northwestern United States. Journal of Climate, 13(11): 1936–1950.
Wu, Z. & Huang, N.E. 2005. Ensemble Empirical Mode Decomposition: A noise-assisted data analysis
method, Centre for Ocean-Land-Atmospheric Studies Tech. Rep. 193, Cent. for Ocean-Land-Atmos.
Stud., Calverton, Md. 1–51 (ftp://grads.iges.org/pub/ctr/ctr_193.pdf ).
Wu, Z., Huang, N.E., Long, S.R. & Peng, C.K. 2007. On the trend, detrending and variability of
nonlinear and non-stationary time series. Proceedings of National Academy of Science USA, 104(38):
14889–14894.
86

Management of beach morphological changes on the

Trivandrum coast
M.U. Niloofar & K.P. Indulekha

Department of Civil Engineering, College of Engineering, Trivandrum, Kerala, India
ABSTRACT: Coastal systems are dynamic environments which comprise three main inter-
related components namely; morphology, sediments and forcing parameters. Numerical
models for shoreline evolution are useful tools in establishing trends and forecasting shoreline
position scenarios for decadal temporal scales. In the present study, an attempt was made to
study beach morphological changes in the Valiathura–Poonthura stretch of the Trivandrum
coast using LITPACK software. Sediment transport, profile change and shoreline evolution
were simulated using different modules of the LITPACK model. The southern-most region
of the stretch is undergoing severe erosion during the monsoon. Coastline evolution for a
period of ten years implies that the coast is undergoing erosion and the beach is declining
eastward. An average approximate decline of 30 m was observed for the shoreline over the
period 2005–2015. A numerical model was applied to analyze the best layout of protective
structures and found that groynes with a spacing of twice the length was effective for sedi-
ment trapping and shoreline development when compared to detached breakwaters.
1 INTRODUCTION
Complex and diverse types of natural processes that occur on the coastal zone bring physi-
cal, chemical and biological changes to fragile coastlines. The coastline of India is undergoing
changes due to several human interventions. Shore erosion is currently causing damage to shore-
lines and public properties, not only along the coast of Kerala, but also around the world. There
are various coastal protection methods like seawalls, groynes, detached breakwaters and so on.
The present study deals with studies on beach morphological changes on the Valiathura–
Poonthura coastal stretch using LITPACK software. Various configurations of breakwaters
and groyne fields were analyzed to identify the best suited structure for coastal protection.
Shoreline change is a complex process which depends upon various factors like tempera-
ture, wind velocity, wave climate, sediment properties, cross-shore profile and so on. Develop-
ment of an accurate model for shoreline change is a difficult process which requires a large
amount of data and a considerable amount of time and effort.
Thach et al. (2007) studied shoreline change by using a LITPACK mathematical model
in Cat Hai Island, Vietnam. According to the simulated and calculated results, the selected
protected construction system, which includes revetments, T-shape sand preventive construc-
tions and submerged breakwaters, is the most suitable and reasonable counter measure for
Cat Hai shoreline stabilization.
Shamji et al. (2010) studied the application of numerical modeling of morphological
changes in a high-energy beach during the south-west monsoon. The LITPROF module of
the LITPACK software was found to accurately simulate beach morphological changes by
adjustment of the calibration parameters. The model performance, computed using different
statistical methods, was found to be good.
Christy Paul and Abilash. P. Pillai (2014) conducted a study on shoreline evolution due to
the construction of rubble-mound breakwaters at Munambam inlet. They found that accre-
tion dominates shoreline changes and there was a net advance in the shoreline. Thus, the break-
waters were found to be very effective in trapping the littoral sediments along the shoreline.
87

Noujas et al. (2014) conducted a study on management of shoreline morphological changes
consequent to breakwater construction. Construction of breakwaters to develop a fishing
harbour at Muthalapozhi had caused erosion immediately north of the inlet and beach build
up south of the inlet. Shoreline changes, nearshore processes and beach characteristics along
this sector were studied through extensive field observations.
2 METHODOLOGY
Coastal processes such as shoreline changes, nearshore waves, long-shore currents and sedi-
ment characteristics were studied to understand various morphological changes including
coastal erosion along the Valiathura coast. The data collected provided the input for numeri-
cal model studies. The input data included wave climate data, initial coastline, initial cross-
shore profile, sediment characteristics and so on. Wave data including wave height, wave
period and wave direction for the period 1981–1984 were collected. As there was no signifi-
cant variation in wave characteristics over these years, the same data was used for simula-
tions for the year 2005. The cross-shore profile and sediment characteristics, defined from
a depth of 5.4 m up to 3.2 m above mean surface level for 2005 was obtained from Shamji
et al (2010). Initial shoreline position was obtained from Google Earth, ArcGIS and MIKE
Zero. A baseline, which is nearly parallel to the coastline, was drawn in the georeferenced
image of the study area. The baseline is 2 km long and it is divided into grids of 100 m
length. The distance from baseline to coastline from each grid point is measured in ArcGIS.
The numerical modeling of shoreline change was done using LITPACK. The governing
equations were solved using a finite difference approach and the software computes wave,
shoaling, refraction, diffraction and the resulting sediment transport at each time step for
each grid point. The governing equation for sediment transport is shown in Equation (1):
d  du  ds
τb −  ρED  = − xy + τ w + τ cur (1)
dy  dy  dy
where τb is bed shear stress due to the long-shore current, ρ is density of water, E is momen-
tum exchange coefficient, D is water depth, u is long-shore current velocity, y is shore-normal
coordinate, sxy is shear component of the radiation stress, and τw & τcur are driving forces due
to wind and coastal current.
The cross-shore profile model describes cross-shore profile changes based on a time series
of wave events. The main assumptions in the profile change model are that long-shore gradi-
ents in hydrodynamic and sediment conditions are negligible and depth contours are parallel
to the coastline. Coastal morphology is described by the cross-shore profile. Wave transfor-
mation across the profile is calculated including the effects of shoaling, refraction, bed fric-
tion and wave breaking. Bed level change is described by a continuity equation for sediment:
∂h 1 ∂ qs
=− (2)
∂t 1− n ∂x
where h is bed level, qs is cross-shore transport and n is porosity. The boundary condition is
that the sediment transport is zero at the coastline. The shore line change model solves the
continuity Equation (2) for the coastline:
∂y c 1 ∂Q Qsou
=− + (3)
∂t h act ∂x hact ∆x
where yc is the distance from the baseline to the coastline, T is time, hact is height of the active
cross-shore profile, Q is long-shore transport of sediment expressed in volumes, x is long-
shore position, ∆x is long-shore discretization step and Qsou is the source/sink term expressed
in volume/∆x.
88

The shoreline along the Valiathura coast is simulated and calibrated with the observed
shoreline data from Google Earth for one year. After the validation of the model, shoreline
evolution for ten years was simulated. The effect of protective structures on shoreline man-
agement was also studied.
3 NUMERICAL MODELING
3.1 Model inputs

The main input to the model is initial coastline, given as a distance from a baseline and the
cross-shore profiles. Historical Google imageries are used to represent the shoreline position.
The cross-shore profile represents the cross-shore variation of depth contours. A grid spac-
ing of 100 m is considered for an accurate representation of the shoreline. The coastline is
determined by drawing the baseline at an angle of 226°N. The initial coastline is illustrated in
Figure 1. Cross-shore profile (Figure 2) values are given at 5 m intervals.
3.2 Wave data

Accurate assessment of the actual wave conditions and their occurrences is extremely impor-
tant for the computation of the littoral drift at the shoreline. Wave climate is defined by the
measured wave parameters like wave heights, wave periods and wave directions, recorded for
the period 1980–1984, which was used for modeling purposes.
The wave data used for the study were derived from the nearshore wave data recorded at
5.5 m water depth at Valiathura. Wave height (Hs), wave period (Tz) and wave direction (°N)
used in this study varies from 0.263–1.889 m, 3.615–10.38 s and 190–270°N, respectively.
3.3 Sediment characteristics

Sediment consists of fine to medium sand with a mean grain size varying from 0.11 to
0.39 mm. This includes sediment from the nearshore and the beach. Sediment parameters
pertaining to sediment characteristics used in the model are fall velocity, bed roughness, spe-
cific gravity and geometrical spreading.
3.4 Sediment transport modeling

The sediment transport modal simulates the cross-shore distribution of wave height, setup
and long-shore current for an arbitrary coastal profile. Rayleigh distribution is used for
wave spectral analysis. LITDRIFT calculates the net/gross littoral transport over a specific
design period and calculates seasonal and annual littoral drift rates. The profile normal
to shoreline was computed as 2260N. The cross-shore profile covers a distance of 180 km
extending up to 5.4 m depth from mean sea level. The profile was discretized with a 5 m
grid size.
Figure 2. Profile series plot of cross-shore

Figure 1. Initial coastline. profile.
89

3.5 Profile change modeling
LITPROF describes the cross-shore profile changes based on a time series of wave events.
Through successive calls to the sediment transport program (STP), the utility program PRFT-
ABL calculates and tabulates transport rates for an envelope of hydrodynamic conditions.
The input files such as cross-shore profile and wave climate files with six items were prepared.
3.6 Shoreline change modeling

Based upon the results from LITDRIFT, LITLINE simulates the coastal response to gradi-
ents in the long-shore sediment transport capacity resulting from natural features and a wide
variety of coastal structures.
Littoral transport was calculated using the LITDRIFT module. It was seen that from June
to August, the beach was eroded with high rate of sediment transport. Maximum erosion
occurred in the month of June and the corresponding sediment transport was 99971.448 m3/
month. Sediment transport during the entire beach building period extending from September
to April, was toward the north with maximum values in September and October. The calcu-
lated net and gross transport were 0.1155 × 105 m3/year and 0.4059 × 105 m3/year, respectively.
The validation of LITPROF was carried out for Valiathura for the peak monsoon period
of mid-June 2005. High erosion was observed during this period as seen in the beach profiles
presented in Figure 3 The berm was completely eroded and deposited as a bar offshore. The
input data for simulations included wave parameters, cross-shore beach positions, berm char-
acteristics and so on. To minimize the difference between measured and computed values,
the model was run for many cases. The main calibration parameter was the scale factor. The
model was run for scale factors of 0.8, 0.9 and 1. The simulated profiles for each scale factor
after 14 days was compared with the actual cross-shore profile (Shamji et al., 2010).
After validating the profile change model, the profile change evolution for short term were sim-
ulated. The predicted profile at the end of a 14-day period from June 4, 2007 is given in Figure 3.
The simulated profile shows high erosion at the beach face and subsequent deposit of sedi-
ment in the offshore region. The simulated profile shows the characteristics of a bar profile,
due to the impact of monsoon waves. Simulation results from LITPROF were compared with
simulation results from the profile change model (Shamji et al., 2010) for the same period,
which is given in Figure 4. It can be seen that the LITPROF model output is quite compara-
ble to the profile change model output.
Shoreline evolution has been modeled using the LITLINE module. The model is validated
for one year. The actual shoreline for 2005 was obtained from Google Earth which was used
as the initial coastline. The shore line after one year was also obtained from Google Earth,
which was used as the final coastline for validation. The input data for simulations included
initial coastline, wave parameters, cross-shore beach positions and so on. In LITLINE, the
Figure 4. Comparison between simulated results

Figure 3. Predicted beach profile change and results from the profile change model.
after 14 days.
90

main calibration parameter was active depth. The model was run for active depths of 3 m,
3.5 m, 4 m, 4.5 m, 5 m and 5.5 m. From the simulations, an active depth of 4 m gave satisfac-
tory results. An active depth of 4 m was used for shoreline evolution.
After validating the LITLINE modal the shoreline evolution for a ten-year period was
simulated. The initial coastline and cross-shore profile for the year 2005 was inputted. Wave
climate data for ten years was prepared by the LITCONV utility in LITPACK. Sediment
transport tables were generated using the LINTABLE utility. Then, the shore line evolution
for ten years was computed. The model was run without considering structures and sources.
The simulated shoreline evolution is given in Figure 5.
From the shoreline evolution it is clear that the shore is continuously eroding over time.
Erosion is more prevalent in the southern-most region. The selected region is undergoing
severe erosion and hence, the beach is declining eastward. The average approximate decline
for the shoreline over the period 2005–2015 was 30 m.
The simulated shoreline for 2015 was compared with the actual observed shoreline of 2015
obtained from Google Earth, as shown in Figure 6.
The simulated shoreline from model gives satisfactory results. The effectiveness of detached
breakwaters and groynes for the protection of the Trivandrum coast in Kerala was numeri-
cally investigated using LITLINE software.
4.1 Effect of detached groynes on shoreline change

Case i. 8 nos with 40 m length (L) & 2 L spacing
The shoreline change from 2015 to 2016 due to the effect of groyne field having 8 numbers
of gryones with 40 m length and 2 L spacing, was analyzed. After installing groynes, little
sediment was trapped within the groyne field. Figure 7 shows the results from LITLINE.
Case iii. 8 nos with 50 m length (L) & 2 L sapcing
The shoreline change from 2015 to 2016, due to the effect of a groyne field with eight
groynes of 50 m length and 2 L spacing, was analyzed. By providing 50 m length, the beach
was stabilized which is clear from Figure 8.
Figure 6. Comparison between observed and

Figure 5. Shore line evolution for ten years. obtained shoreline change of 2015.
Figure 7. Shore line change due to effect of Figure 8. Shore line change due to effect of
40 m length groyne field with 2 L spacing. 50 m length groyne field with 2 L spacing.
91

Case iv. 8 nos with 50 m length (L) & 2.6 L spacing.
The shoreline change from 2015 to 2016, due to the effect of a groyne field with eight
groynes of 50 m length and 2.6 L spacing, was analyzed. Figure 9 gives the results from
LITLINE.
From the above results, a groyne field with eightgroynes of 50 m length and 2 L spacing
was found to be effective for stabilizing the beach.
4.2 Effect of detached breakwater on shoreline change

Case i. 3 nos 200 m length, 100 m spacing & 100 m from baseline
The shoreline change from 2015 to 2016, due to the effect of a detached breakwater with
three breakwaters of 200 m length and 100 m spacing, at a distance of 100 m from baseline,
was analyzed. Figure 10 gives the results from LITLINE.
Case ii. 3 nos 100 m length, 200 m spacing & 80 m from baseline
three breakwaters of 100 m length and 200 m spacing, at a distance of 80 m from baseline,
was analyzed. Figure 11 gives the results from LITLINE.
Case iii. 3 nos 100 m length, 200 m spacing & 100 m from baseline
three break waters of 100 m length, 200 m spacing, at a distance of 100 m from baseline, was
analyzed. Figure 12 gives the results from LITLINE.
From the study, it is clear that groyne fields of 50 m length and 2 L spacing trapped maxi-
mum sediment in the up-drift side when compared to a detached breakwater. No significant
improvement in shoreline due to installation of breakwaters and also results in down drift ero-
sion. It is clear that increased spacing between the groynes resulted in poor retention and scour
within the groyne field. Groynes placed with a spacing of twice the length are a better solution.
Figure 10. Shore line change due to effect of

Figure 9. Shore line change due to effect of 50 m 200 m length, 100 m spaced detached break water,
length groyne field with 2.6 L spacing. 100 m from baseline.
Figure 11. Shore line change due to effect of Figure 12. Shore line change due to effect of
100 m length, 200 m spaced detached breakwater 100 m length, 200 m spaced detached breakwater
at 80 m from baseline. at 100 m from baseline.
92

5 CONCLUSION
It was concluded that the selected region undergoes minor accretion for the first five months
of the year. For the remaining six and a half months, the coast undergoes severe erosion and
the maximum erosion occurs in the month of June. The calculated net and gross transport
are 0.1155 × 105 m3/year and 0.4059 × 105 m3/year. During the monsoon it was found that the
beach profile undergoes a high erosion at the beach face, with subsequent deposit of sedi-
ment at an offshore region resulting in bar formation.
Coastline evolution for a period of ten years of the Valiathura region implies that the
coast is undergoing severe erosion and hence, the beach is declining eastward. An average
approximate decline of 30 m was observed for the shoreline over the period 2005–2015. The
southern-most region of the coast is subject to severe erosion. It was found that groynes offer
a better solution for the coastline erosion of the Trivandrum coast with respect to sediment
trapping efficiency as well as shoreline development when compared to breakwaters. It was
also observed that spacing between groynes equal to twice the length is a better option com-
pared to spacing of 2.6 times the length.
REFERENCES
Noujas, V., Thomas, K.V., Sheela Nair, L.S., Hameed, T.S., Badarees, K.O. & Ajeesh N.R. (2014). Man-
agement of shoreline morphological changes consequent to breakwater construction. Indian Journal
of Geo-Marine Sciences, 43(1), 54–61.
Paul, C., Pillai. C.A. (2014). Shoreline evolution due to construction of rubblemound breakwaters at
Munambam inlet. International Conference on Innovations & Advances in Science, Engineering and
Technology [IC—IASET 2014], 3(5), 462–467.
Shamji, V.R., Kurian, N.P., Thomas, K.V. & Shahul Hameed, T.S. (2010). Application of numerical
modelling for morphological changes in ahigh-energy beach during the south-west monsoon. Current
Science, 98(5), 691–695.
Thach, N.N., Truc, N.N. & Hau, L.P. (2007). Studying shoreline change by using LITPACK mathemati-
cal model (case study in Cat Hai Island, Hai Phong City, Vietnam.VNU Journal of Science, Earth
Sciences, 23, 244–252.
93

Impact of climate change on stream flow
P. Dhanya
WRHI, Government Engineering College, Thrissur, India
Reeba Thomas
Department of Civil Engineering, Government Engineering College, Thrissur, India
ABSTRACT: River discharge is a critical component of the global water cycle and much
concern has been raised regarding the changes it has experienced in recent years due to
unprecedented changes in temperature and precipitation. The Soil and Water Assessment
Tool (SWAT) was used for hydrologic modeling of the Thoothapuzha basin to compute the
effect of climate change on stream flow by simulating the model using dynamically down-
scaled climate data. Future changes in climate were assessed using Global Climate Mod-
els (GCMs). A dynamic downscaling model was applied to reduce large-scale atmospheric
variables in the GCMs into localized weather variables. Climate data were dynamically
downscaled using a Regional Climate Model (RCM) REMO2009 under the driving GCM,
The Max Planck Institute for meteorology Earth System Model-Low Resolution (MPT-
ESMLR) for a future period of 2016–2030. An emission scenario of RCP 4.5 was consid-
ered. A delta change method of bias correction was applied to the downscaled data in order
to get the corrected daily weather data. The model was calibrated and validated and the
performances evaluated using historic weather data (1974–2012) in terms of R2, Percentage
Bias (PBIAS), Nash-Sutcliffe Efficiency (NSE), and the values obtained were 0.79, 6.829 and
0.78, respectively, which shows that this basin could very well be modeled in SWAT. Stream
flow forecasting was carried out with bias corrected RCM data in a calibrated SWAT model.
The decreasing trend in the forecasted stream flow for the period of 2016–2030 under RCP
4.5 shows the impact of increased greenhouse gas emissions and the necessity of remedial
measures to be taken in order to preserve our water resources for the coming generations.
1 INTRODUCTION
Climate change is expected to alter the timing and magnitude of runoff, which has significant
implications for current and future water resource planning and management. A rise in mean
global temperature due to increased CO2 emissions indicates a major human induced climate
change which can cause alteration in the hydrologic cycle. This phenomenon has had discern-
ible impacts on the physical, ecological and biological systems of the earth. Climate change
is expected to adversely impact water resources, water quality and freshwater ecology. The
evolution of future greenhouse gases is highly uncertain. The Special Report on Emission
Scenarios (SRES) by Inter governmental Panel on Climate Change (IPCC) establishesdif-
ferent future world development possibilities and the corresponding CO2 emissions in the
twenty-first century, taking into consideration the possible changes in various factors includ-
ing economic development, technological development, energy intensities, energy demand,
and structure of energy use, resources availability, population change and landuse change.
Global Climate Models (GCMs) have evolved from the Atmospheric General Circulation
Models (AGCMs) widely used for daily weather prediction. This is widely used to forecast
future weather parameters under the influence of the increasing atmospheric carbon dioxide
(CO2) predicted by IPCC. Most GCMs neither incorporate nor provide information on scales
smaller than a few hundred kilometers. It is possible to model small scale interactions and
95

establish relationships between present-day local climate and atmospheric conditions through
the downscaling process, so that information is more realistic at a finer scale, capturing local
heterogeneities. This downscaling can be of two types, dynamical or statistical. Dynami-
cal downscaling relies on the use of a regional climate model (RCM), with high resolution.
RCMs use the large-scale atmospheric information supplied by GCM output at the lateral
boundaries and incorporates more complex topography, in order to generate realistic climate
information at a spatial resolution of approximately 20–50 kilometers. Statistical downscal-
ing involves the establishment of empirical relationships between historical and/or current
large-scale atmospheric and local climate variables and the same relationship is assumed to
exist between present large-scale variables and local variable under possible future climate
conditions (Manu et al., 2011). The impact of changing climate conditions on the hydrologic
cycle can be analyzed by integrating a global climate model with a hydrologic model (Negash
et al., 2013). In this study, a global climate model was used to produce the climate data such
as temperature and precipitation which act as boundary conditions to run the RCM. The
data taken from the global climate model were then extracted to the basin area using multi
dimension tools in ArcGIS. A bias factor was applied to these extracted data in order to
bring the statistical parameters (mean, standard deviation, etc.) of both observed and future
forecasted climate variables within comparable limits. Then, these data were incorporated
into the hydrological model to obtain the desired results (Johnson et al., 2012).
Nomenclature
AGCM Atmospheric General Circulation Model

CFSR Climate Forecast System Reanalysis data
GCM Global Climate Model
GIS Geospatial Information System
HRU Hydrologic Response Unit
IPCC Intergovernmental Panel on Climate
Change
NCEP National Center for Environmental
Prediction
NSE Nash Sutcliff Coefficient
PBIAS Percentage Bias
RCM Regional Climate Model
RCP Representative Concentration Pathways
REMO Regional Model
SRES Special Report on Emission Scenario
SWAT Soil and Water assessment Tool
2 OBJECTIVE OF THE STUDY
The main objective of this study was to assess the impact of climate change on stream flow
in the Thoothapuzha basin by conducting hydrologic modeling in SWAT. For this, model
calibration and validation were first carried out using historical daily weather and rainfall
data (1974–2014). Bias corrected dynamically downscaled GCM weather data were used to
forecast if the future runoff in Thoothapuzha corresponds to emission scenario RCP 4.5 and
the change in runoff was analyzed.
3 STUDY AREA AND DATA REQUIRED
This research focuses on Thuthapuzha, one of the tributaries of Bharathapuzha, which

is located 10°30′00″–11°40′00″ N and 76°10′00″–77°0′0″ E. It receives an average annual
rainfall of 2,683 mm. The average maximum temperature of the subbasin is 31.43°C,
with a minimum temperature of 24.33°C. Thuthapuzha Sub Basin is located toward the
96

northwestern side of Bharathapuzha. It covers a total area of 1018 km2. Of the total
area, 75% falls within the Palakkaad district and the remaining 25% in the Malappuram
district. Approximately 30% of the basin area is forest and 25% is agricultural land. In
total, 1.2% of the area consists of water bodies. Silent Valley National Park is situated in
this catchment. Data required for this study include the following, SRTM data, landuse
map, soil map, weather data, rainfall data and discharge data. Daily weather data were
taken from records of the SWAT weather database organized by the National Center for
Environmental Prediction (NCEP), Climate System Reanalysis data, for six stations for
the period 1979–2014. In this study, weather data for future forecasting were dynamically
downscaled using the RCM REMO 2009, under the GCMMPT-ESM-LR downloaded
from the Coordinated Regional Climate Data Downscaling Experiment (CORDEX) for
the period 2016–2030.
4 METHODOLOGY
4.1 Watershed modeling

Watershed modeling was carried out in SWAT, a physical model, which works on a set of
conservation equations which govern nature. Delineation of the Thoothapuzha basin was
done by considering the Pulamanthole gage station as the outlet point. The subbasins were
again divided into a number of Homogeneous Response Units (HRU) having unique soil,
land and slope classes. Daily weather data including maximum and minimum temperature,
wind, solar radiation, humidity obtained from NCEP CFSR data for the period 1979–2014
were used. The model was then calibrated (1994–2000) and validated (2001–2012) using
observed monthly discharge data. The model performance was evaluated in terms of NSE,
PBIAS and R2 coefficients.
4.2 Climate data extraction

Future climate data (2016–2030) forecast with GCM MPT-ESM-LR corresponds to emis-
sion scenario B1 and RCP4.5; downscaled dynamically with RCM REMO 2009 (Amin et al.,
2015) and was downloaded from the CORDEX South Asia data portal (Center for Climate
Change Research cccr.tropmet.res.in/home/ftp_data.jsp) and extracted to the centroid of the
basin by using a multidimensional tool in ArcGIS to accommodate the spatial changes in
the weather pattern. A delta change method of bias correction was performed to ensure that
important statistics of the downscaled output matches with the corresponding statistics of
the observed data:
P*cont(d) = Pobs(d) (1)
P*scen(d) = Pobs(d) × (µm(Pscen(d)))/(µm(Pcont(d))) (2)
T*scen(d) = Tobs(d) (3)
T*scen(d) = Tobs(d) + [µm(Tscen(d))–µm(Tcont(d))] (4)
where, P*cont(d) is the final bias corrected precipitation in the control period, Pcont(d) is
the precipitation in the control period, Pobs(d) is the observed daily precipitation, µm, mean
within monthly interval, Pscen(d) is the precipitation in future scenario and P*scen(d) is the
final corrected precipitation in the future scenario.
4.3 Runoff forecasting

The SWAT weather parameters were calculated from the bias corrected GCM data and the
calibrated SWAT model used for runoff forecasting. This was done for the period 2016–2030
for selected emission scenario B1 and RCP 4.5. Seasonal change in runoff pattern was then
analyzed (S.L. Neithish et al.).
97

5.1 Watershed modeling

First, the watershed was delineated taking the Pulamanthole gage station as the outlet point.
The delineated watershed is shown in Figure 1(a) and the watershed modeling is carried
out. The graph showing calibration (1994–2000) and validation of the model is shown in
Figures 1(b) and 1(c), respectively.
Model performance was analyzed in terms of Nat-Sutcliffe Efficiency (NSE), percentage
bias (PBIAS), coefficient of determination (R2). Values obtained during calibration and vali-
dation are shown in Table 1.
5.2 Trend analysis of weather data

The long-term trend in the dynamically downscaled weather data (mainly temperature
and precipitation) for emission scenario RCP 4.5 was analyzed graphically for the period
Figure 1(a). Delineated watershed of Thuthapuzha.
Figure 1(b). Model calibration.

Figure 1(c). Model validation.
Table 1. Coefficient values in calibration and validation.
Coefficient Calibration (1994–2000) Validation (2001–2012)
NSE 0.78 0.75

PBIAS 6.82 9.89
R2 0.79 0.76
98

2016–2030. It was seen that there will be a considerable decrease in precipitation by the year
2030. In order to verify the genuineness of this trend, the historic data were also analyzed,
yielding the same decreasing trend (Figure 2(b)). The rate of decrease was milder in the
historical period in contrast to what is going to happen in the future under RCP4.5. Simi-
larly, the trend analysis on minimum and maximum temperature showed a slight increase
in its value under RCP 4.5. The results obtained, indicating the annual trend, are shown in
Figures 2(a), 2(b) and 2(c).
5.3 Prediction of future runoff

The validated SWAT model was used to predict the future runoff for which future climato-
logic data obtained by dynamical downscaling of global weather data, bias corrected by a
delta change method, were used. Runoff volume was analyzed by converting it into cumu-
lative volume in monsoon and summer seasons. Figures 3(a) and 3(b) show the discharge
during monsoon (June–October) and summer (remaining period) seasons forecasted for the
period 2016–2030 for RCP 4.5, respectively. Discharge during both monsoon and summer
show a decreasing trend with irregular peaks, which indicate a major water scarcity in future
decades. From the graph, it is clear that in 2030, the monsoon discharge will be only 65% of
that available in 2016.
Figure 2(a). Trend analysis of annual precipitation (mm) 2016–2030.
Figure 2(b). Trend analysis of annual precipitation (mm) 1979–2013.
Figure 2(c). Trend analysis of temperature (°C).
99

Figure 3(a). Monsoon discharge in Thoothapuzha (MCM).
Figure 3(b). Summer discharge in Thoothapuzha (MCM).
6 CONCLUSION
Rainfall runoff modeling was successfully calibrated and validated using a SWAT model for
the Thoothapuzha basin. The simulated flow at the basin outlet was compared with the
observed flows in order to carry out calibration and validation of the model. The obtained
values for NSE, R2, PBIAS were satisfactory and within the range. Thus, hydrologic mod-
eling of the Thoothapuzha basin for the period 1979–2014 was successfully completed, with
NSE, PBIAS and R² values of 0.78, 6.89 and 0.79, respectively. Stream flow in the Thooth-
apuzha basin was predicted by using dynamically downscaled climate data for emission sce-
nario RCP 4.5, which is characterized by a medium rate of expected CO2 emission. In this
scenario, the trend analysis of forecasted discharge values shows a decreasing trend in their
magnitude in both monsoon and summer seasons, which can be considered as a warning sig-
nal of the impending reduction in runoff in the coming years and the need to take remedial
measures to address this.
REFERENCES
Amin, M.Z.M., Shaaban, A.J., Ohara, N., Kavvas, M.L., Chen, Z.Q., Kure, S. & Jang, S. (2015). Climate
change assessment of water resources in Sabah and Sarawak, Malaysia, based on dynamically-down-
scaled GCM projections using a regional hydro climate model. Journal of Hydrologic Engineering,
21(1), 05015015(9).
IPCC SPECIAL REPORT EMISSIONS SCENARIOS (2000). ISBN: 92-9169-113-5.
Johnson, T.E., Butcher, T.B., Parker, A. & Weaver C.P. (2012). Investigating the sensitivity of U.S. stream
flow and water quality to climate change: U.S. EPA Global Change Research Program’s 20 Water-
sheds Project. Journal of Water Resources Planning and Management, 138(5).
Manu Sharma, PaulinCoulibaly and Yonas Dibike (2011). Assessing the need for downscaling RCM
data for hydrologic impact study. American Society of Civil Engineers, 16(6), 534–539.
Negash Wagesho, Jain, M.K. & Goel, N.K. (2013). Effect of climate change on runoff generation: Appli-
cation to rift valley lakes basin of Ethiopia. American Society of Civil Engineers, 18(8), 1048–1063.
Neithisch, S.L., J.G. Arnold, J.R. Kiniry, R. Srinivasan & J.R. Williams (2002). SWAT User guide
(2000). Centre for Climate Change Research cccr.tropmet.res.in/home/ftp_data.jsp.
100

Fluid structure interaction in transient cavitating flow

in pipes—a review
R. Miji Cherian, N. Sajikumar & K.S. Sumam

Department of Civil Engineering, Government Engineering College, Trichur, India
ABSTRACT: Transient flow is the intermediate stage flow when the flow conditions are
changed from one steady state to another, caused by the sudden changes in the operating
conditions, such as abrupt closure or opening of the control valve and starting or stopping
of pumps. In liquid-filled pipe systems, such disturbances create pressure waves of large
magnitude which travel to and fro. The propagation and reflection of these pressure waves
is referred to as hydraulic transients or ‘water hammer’, which can either rise or drop the
normal working pressure of the pipe system. Cavitation (which is the formation, growth and
collapse of vapor bubbles in a flowing liquid) can result in a region when the pressure of the
liquid drops below its vapor pressure. Pipe systems also experience severe dynamic forces dur-
ing transient flow due to fluid structure interaction. Fluid induced structural motion, struc-
ture induced fluid motion, and the underlying coupling mechanisms, are commonly referred
to as Fluid Structure Interaction (FSI). This study aims to review a sample of the relevant lit-
erature on the main related topics, such as water hammer, cavitation, FSI and viscoelasticity.
1 INTRODUCTION
For any piping system, sudden operating conditions, such as closing or opening of the valve,
and starting and stopping of the pump, will induce transient flow. Subsequently, it causes
‘water hammer’, cavitation, Fluid Structure Interaction (FSI), and energy dissipation due
to viscoelasticity. A general review of these phenomena is discussed in the following sections
which include numerical and experimental investigations carried out over the past decades.
2 WATER HAMMER
Abrupt changes in the flow condition, such as closing or opening of the valve, and starting or
stopping of the pumps, are unavoidable situations in hydraulic systems, and are ones which
cause water hammer. The present water hammer theory is based on the Joukowsky formula,
derived in the second half of 19th century. As per this formula, a change in pressure depends
on a change in velocity, mass density of the fluid and the velocity of sound in the fluid. Water
hammer equations include the continuity equation and the equation of motion. Streeter and
Wylie (1967) introduced the Method of Characteristics (MOC) to solve these equations numer-
ically. Chaudhry and Hussaini (1985) presented three explicit finite difference schemes: Mac-
Cormac, Lambda and Gabutti for the analysis of transient flow through pipes. All these three
schemes are second-order accurate both in space and time and have a predictor step and a
corrector step. They conducted numerical analyses for studying the effect of courant number
on the effectiveness of the methods, and found that when the courant number becomes one,
there is no advantage for second-order accurate schemes over first order accurate schemes.
Hadj-Taıeb and Lili (2000) presented another mathematical formulation for the analysis of
gas-liquid mixture in deformable pipes during water hammer. Two different methods were used
for the analysis: MOC and the finite difference conservative method. The model developed was
101

validated using test results available from a literature review, and found that the finite difference
conservative method had close agreement with the test results.
Brunone et al. (2000) conducted a number of experiments to investigate the effect of veloc-
ity profiles and unsteady pipe friction in transient flow. Experiments were conducted on a
polyethylene pipe for different flow rates. A new mathematical model was developed using
the MOC which considered unsteady friction. The experimental results were compared with
classical water hammer equations and also with the new model, and found that the new
model performed reasonably well with the experimental results. Adamkowski (2001) inves-
tigated the penstock failure at Lapino Power Plant and found that, after a rapid flow cut-
off, an excessive pressure rise occurred because of the presence water hammer, which then
resulted in the bursting of a pipe. This study showed the importance of rate of flow cut-off,
and the associated maximum pressure rise, in the design of hydropower plants.
Greyvenstein (2002) presented an implicit pressure correction method incorporating a
time step weighing factor, which can be used for liquid and gas flows, steady and transient
flows, and isothermal and non-isothermal flows. The numerical scheme developed was com-
pared with the Lax-Wendroff method and the MOC, and it was found that for the optimum
value of the time step weighing factor, the developed method had good agreement with the
second-order two-step Lax-Wendroff method. By the integration of the Euler equation for
three dimensional domains, Gerbi (2002) developed a numerical model for compressible flow
in deformable pipes. The model was validated with the results of a water hammer test, and
he concluded that the pressure developed during water hammer is less in deformable pipes,
which absorb a great part of the energy.
Kaliatka and Vaisnoras (2005) simulated the water hammer phenomenon using a RELAP5
computer code and compared the results with available test results. They found that the first
pressure peak matched well with the test results, but as time progressed, the accompanying pres-
sure peaks deviated slightly. A historical review by Bergant et al. (2006) included a detailed survey
of the numerical models and experimental work conducted in the field during the 20th century.
To reduce the computational cost for simulating pipe entrance flow, Shen et al. (2006)
introduced a new model using the multigrid finite-volume technique. In this model, the equa-
tions were first solved using a coarse grid and the results were used for a better initial guess
taken from selecting the next finer grid. This process was repeated until the required accu-
racy was achieved. Mitosek and Szymkiewicz (2012) conducted a detailed experimental and
numerical study of unsteady pipe flow. A pressurized tank-pipe-valve system was used for
the study. For improving the numerical model and to reduce the discrepancy noticed between
experimental and numerical models, the concepts of unsteady friction, viscoelasticity and
variable wave celerity were included. For smoothening the pressure wave, a diffusive term
was also introduced in the mathematical model which found that the model performance was
close to the experimental results.
Wilcox (2012) conducted fluid transient analysis for the partial closing of a valve using AFT
impulse software. Amara et al. (2013) modified the finite difference MacCormack scheme by
introducing an artificial viscosity and dissipation constant to minimize the dissipation errors.
The two-step Runga Kutta splitting technique was used for solving the governing equations,
and this adapted the MacCormack scheme, giving better results compared to the original one.
Abuiziah et al. (2013) presented a numerical model for the analysis of transient flow, taking into
consideration the influence of location of surge protection devices and valve closure time on
pressure peak. From the study, it was found that a surge tank installed immediately upstream of
the butterfly valve considerably reduced the effect of overpressure and low pressure.
The effect of pipe roughness and reservoir water levels on the pressure waves during tran-
sient flow, was investigated by Mansuri et al. (2014) and found that pressure head decreased
with an increase in pipe roughness and increased with reservoir water level. Monajitha et al.
(2014) modified the Mitosek and Szymkiewicz (2012) model by accounting for artificial vis-
cosity for reducing the dispersive error, and found that the model results agreed well with
previous published results, than that obtained from surge 2000 software. MOC and the finite
difference MacCormack scheme are widely used in the analysis of water hammer. Governing
equations of water hammer are used with certain modifications, such as unsteady friction,
102

viscoelasticity and variable celerity. Artificial viscosity was introduced to the numerical meth-
ods to reduce the discrepancy between experimental and numerical results.
3 CAVITATION
Water hammer waves cause an alternating pressure rise/drop in a pipe system. When low pres-
sure prevails for a long time, the system is subjected to cavitation. Research on cavitation
covers a very wide area and includes gaseous cavitation, vaporous cavitation and column sepa-
ration. Among these, transient vaporous cavitation is mainly looked at in this study. Transient
vaporous cavitation is the formation, growth and collapse of vapor bubbles in a flowing liquid
in a closed conduit, in a region where, the pressure of the liquid drops below its vapor pres-
sure. Simpson and Bergant (1994a) summarized the experimental investigation in the field of
water hammer with column separation. Experiments were conducted in a copper pipeline with
two flow visualization polycarbonate blocks. Data acquisition systems consisted of pressure
transducers, a velocity measurement system and temperature measurement using a hot film
thermal probe. A computer data acquisition system was used for continuous recording and
storing of data. Simpson and Bergant (1994b) compared six cavity models including the Dis-
crete Vapor Cavity Model (DVCM) and the Discrete Gas Cavity Model (DGCM). The effect
of input parameters, such as wave speed, friction factor, initial velocity, pipe diameter and pipe
length, on the maximum pressure rise was also included in this investigation.
An analytical study on transient vaporous cavitation was conducted by Singhal et al.
(2002). In order to compute the change in fluid property during phase transition the vapour
transport equation was considered. The vapour mass fraction is governed by the transport
equation. The bubble dynamics expressions and phase change rate expressions were also
derived in this study. Shu (2003) developed a two-phase homogeneous equilibrium vaporous
cavitation model and compared it with the conventional column separation model in which
bubble dynamics was not a factor. Mass transfer during cavitation was ignored in this model,
but frequency dependent friction was included. The developed model predicted the pressure
peaks and dissipation more accurately.
A historical review by Bergant et al. (2006) included a detailed survey of the numerical
models and experimental work conducted in the field of water hammer with column sepa-
ration during the 20th century. Urbanowicz and Zarzycki (2008) compared four cavitation
models for transient cavitating flow in pipes with the experimental results. The models con-
sidered were the column separation model, Adamkowski’s model, the gas vapor cavitation
model and the bubble cavitation model. Among these four models, the column separation
model, Adamkowski’s model and the bubble cavitation model were very good in the simula-
tion of cavitation, and Adamkowski’s model found to be the best model among the three
models which correctly simulates pressure amplitude.
Adamkowski and Lewandowski (2009) developed a new DVCM and compared it with
the previous DVCM model. In the new model, MOC was used for analysis by incorporat-
ing unsteady friction. Experiments were conducted in a copper pipeline with two visualiz-
ing segments made of plexi-glass and it was found that vaporous cavitation was distributed
along the pipeline, with maximum concentration close to the valve, which decreased with an
increase in distance from the valve. Sumam et al. (2010) presented an alternate approach for
modeling transient vaporous cavitation. Cavitating flow was simulated using continuity and
momentum equations for water vapour mixtures, and the transport equation was used for the
vapour phase. The MacCormack scheme was used for developing the model. Model results
matched well with the published and experimental results.
4 FLUID STRUCTURE INTERACTION (FSI)
In conventional water hammer analysis, the effect of pipe material was considered by includ-
ing pipe-wall elasticity. For rigid pipes, this leads to acceptable results, but for flexible
103

systems, the interaction of fluid with structure is greater. In such cases, dynamic analysis of
both fluid and structure has to be done together. Fluid induced structural motion, structure
induced fluid motion and the underlying coupling mechanisms, are commonly referred to
as FSI. The fundamental theoretical basis for FSI in straight liquid-filled pipe was proposed
initially by Skalak. In the studies on FSI, three support conditions are standard: (i) pipe
anchored with expansion joints throughout its length; (ii) pipe anchored throughout against
axial motion; and (iii) pipe anchored at its upstream end. Further to this, three coupling
mechanisms were considered to incorporate the FSI: junction coupling, Poisson coupling
and friction coupling.
Wiggert et al. (1987) analyzed the transient pipe flow with FSI. Axial motion, shear and
bending in the y-z plane and x-z plane and torsion in the pipe wall, for example, were consid-
ered and all the governing equations were solved by the MOC. The numerical model formed
was compared with the experimental work conducted on a system of copper pipes orthog-
onally connected in series. Lavooij and Tijsseling (1991) modeled the FSI using extended
water hammer theory for the fluid and beam theory for the structure. All the three coupling
mechanisms were considered in the model. Two different numerical procedures were used for
solving FSI problems. The first method is a combination of MOC for fluid equations and
the Finite Element Method (FEM for pipe equations).The second is MOC for both fluid
and pipe equations. The effect of valve closure time with FSI was also investigated and it
was found that if the effective closure time of the valve is less than both the Eigen period
of the structure and main timescale of the water hammer waves, FSI is important and to be
accounted.
Heinsbroek & Tijsseling (1994) experimentally investigated the validity of the classical
water hammer analysis for flexible support. Experimental set up consists of seven straight
pipes connected by six 90° bends suspended by steel cables and supported by springs at the
bends. From the experimental and numerical studies they found that for anchor, with rigid-
ity more than the axial stiffness of one meter of pipe, the influence of support anchors on
the transient behavior of the system is small. But for flexible support system, FSI is more
important and to be considered for the determination of equivalent stress in the pipe system.
Tijsseling (1996) reviewed the analytical, numerical and experimental studies conducted all
over the world in the area of FSI; all the governing equations in this field and in numerical
models, such as the 4-equation model and the 14-equation model, were presented in this
study. Three liquid pipe interaction mechanisms of friction coupling, Poisson coupling and
junction coupling, were also included.
Heinsbroek (1997) used two different numerical procedures for solving FSI problems. The
first method was a combination of MOC-FEM and the second was MOC for both fluid
and pipe equations. By comparing the results, it was found that the Bernoulli-Euler theory,
along with MOC-FEM, was the numerical procedure best suited for FSI problems. Tijsseling
(1997) identified a new phenomenon called Poisson coupling beat, due to precursor waves,
which is the cause of the pressure peak greater than that calculated by Joukowsky equation.
The Poisson coupling beat was included in the calculation as a ratio of stress wave speed and
pressure wave speed. The proposed relation connecting the beat period and ratio of wave
speeds was validated numerically in this study.
Tijsseling and Heinsbroek (1999) investigated the influence of bend motion by conduct-
ing a series of tests on a 3D suspended pipeline with a total of six bends—restrained and
unrestrained. Pressure head history was plotted for 16 cases by changing the conditions of
the bend, and found that pressure peak exceeded the value calculated by the Joukowsky equa-
tion. Maximum stress in the pipeline was found to be lowest when most of the bends were
restrained. Wiggert and Tijsseling (2001) conducted a detailed review on the essential mecha-
nism that causes FSI in liquid-filled flexible piping systems.
Li et al. (2003) proposed a new analytical model for the exact solution of the FSI 4-equation
model, based on the D’Alembert wave solution. The proposed model was validated with
the test results of a suspended closed tube set up by Vardy and Fan (1989) and the error
in numerical methods eliminated. Tijsseling and Vardy (2004) distinguished seven types of
fluid-flow with a timescale ranging from infinity to zero, and presented governing equations
104

for all of them. The importance of friction, elasticity, inertia and Poisson’s coupling at differ-
ent timescales were investigated using a numerical model.
To study the effect of FSI and cavitation, Tijsseling and Vardy (2005a) conducted experi-
ments in a T pipe suspended horizontally by steel wires and was excited through the external
impact of a steel rod. A review of different experimental set ups used at the University of
Dundee, UK, for the experimental investigations in FSI is given by Tijsseling and Vardy
(2005b). In a different experimental set up, a single straight pipe, an L pipe, a T pipe and a 3D
pipe system were used. Tijsseling (2007) developed a FSI 4-equation model for the analysis of
thick walled pipes by averaging hoop and radial stresses. The liquid and pipe equations were
coupled through boundary conditions and the numerical model was validated by conducting
experiments on a closed steel straight rod excited by the impact of a solid rod at one end.
Etlender et al. (2007) presented an FSI model for transient flow in flexible hoses. Along
with conservation of mass and momentum, wall shear stress was also included in the model.
The finite-volume method was used for the discretization of fluid equations and the FEM
was used for discretization of pipe structures as a thin shell. An iterative staged algorithm
was used for FSI coupling. Ahmad and Ali (2008) investigated the effect of junction coupling
at valves and junctions of the piping system. MOC-FEM methods were used for the analysis
of fluid and pipe equations respectively. Numerical solution was obtained using MATLAB
and the results showed that junction coupling effects depended mainly on the rigidity of the
piping system.
Tijsseling et al. (2008) reviewed the work of Richard Skalak on wave propagation in an
infinitely long tube filled with fluid. The outcome of Skalak’s work is now used for determin-
ing the dispersion of pressure waves in a liquid-filled long pipe. Along with the classical water
hammer theory, radial inertia of fluid and pipe, and the bending stress and rotary inertia of
the pipe wall were considered for the study. Skalak’s 4-equation model was considered for
the study and it was found that in unrestrained steel and plastic pipes, the length of water
hammer wave front spread due to FSI, was only around ten pipe diameters. Henclik (2010)
presented a new solution algorithm for the 4-equation model and used this for studying a
physical model of a reservoir-pipe-valve system with a number of rigid supports. A pressure
time graph was plotted at the valve end, without inner supports and with inner supports.
Numerical investigations proved that FSI has an important role even for rigid support sys-
tems. Wilcox (2012) conducted transient analysis for partial closing of the valve. In this study,
the fluid transient analysis was done using AFT impulse software from which the results were
exported to CEASAR II software, which performed the transient mechanical analysis of the
piping system.
5 VISCOELASTICITY
Viscoelasticity is the property of materials that exhibit both viscous and elastic character-
istics during deformation. Under applied load, these materials exhibit instantaneous elastic
strain followed by gradual retarded strain. Most of the polymers are viscoelastic in nature,
but in the hydraulic transient analysis, this factor is neglected by most of the researchers.
Viscoelastic behavior of pipe material causes damping of the pressure wave and increases
the rate of energy dissipation. Covas et al. (2004, 2005) presented a mathematical model
with a viscoelastic pipe system and experimentally validated for hydraulic transients. Experi-
ments were conducted in a polyethylene pipe system with data acquisition systems, such as an
acquisition board, pressure transducers, strain gage and a computer. Mathematical models
included unsteady friction and pipe-wall viscoelasticity. For modeling unsteady friction, head
loss was decomposed to a steady state component and unsteady state component. The linear
viscoelasticity model was used for modeling viscoelasticity, which included instantaneous
strain and gradual retarded strain.
Soares et al. (2008) conducted an experimental and numerical analysis of a PVC pipe dur-
ing hydraulic transient. A hydraulic transient solver had been developed for PVC pipes, in
which MOC was used for solving the governing equations. The Kelvin-Voigt mechanical
105

model was used for incorporating viscoelastic behavior. The model was found to be very
effective in predicting both dispersion and attenuation effects during water hammer. They
also identified that the effect of unsteady friction was less than pipe-wall viscoelasticity in
PVC pipes. Soares et al. (2009) investigated the unsteady flow with cavitation in HDPE pipes.
A hydraulic transient solver was developed for HDPE pipes in the same way as Soares et al.
(2008). This solver was calibrated using experimental data by closing a downstream side
valve. For cavitating flow analysis, another set of experiments were conducted by closing
an upstream valve. For the numerical analysis of cavitating flow, three different models were
considered: DVCM, DGCM and the Borga et al. (2004) model. On comparison it was found
that Borga et al. (2004) model agreed well with experimental data.
Keramat et al. (2012) considered FSI and viscoelasticity together for the analysis of tran-
sients. Viscoelasticity was modeled using the Kelvin-Voigt model. The FSI 4-equation model
was derived, including viscoelasticity, and solved numerically using the MOC-FEM approach
and full MOC approach. Numerical results were compared satisfactorily with available exact
results and experimental results.
6 CONCLUSION
There are various operational problems in pipe systems that cause water hammer, cavitation,
FSI and energy dissipation due to viscoelasticity. The review of the individual cases in this
study helps in the understanding of each of them separately, so that they can be considered
together in the analysis. To analyze fluid structure interaction in transient flow, the 4-equation
model considered two fluid equations (continuity and momentum equations) and two struc-
ture equations (for axial and flexural motion of the pipe). Similarly, for modeling transient
vaporous cavitation, the vapour transport equation is most widely used. However, for solv-
ing the governing equations of water hammer, cavitation, fluid structure and interaction the
finite difference method is used for fluid equations and the FEM is used for structure equa-
tions. From this review it was found that a combination of finite difference method and FEM
can be used effectively for the analysis of FSI in transient cavitating flow through pipes.
REFERENCES
Adamkowski A. (2001). Case study: Lapino power plant penstock failure Journal of Hydraulic Engi-
neering, 127(7), 547–555.
Adamkowski A. & M. Lewandowski (2009, October). Improved numerical modelling of hydraulic tran-
sients in pipelines with column separation. In 3rd IAHR International Meeting of the Workgroup on
Cavitation and Dynamic Problems in Hydraulic Machinery and Systems (pp. 419–431).
Ahmad A. & R.K. Ali (2008, October). Investigation of the junction coupling due to various types of
the discrete points in a piping system. in the 12th International Conference of International Associa-
tion for Computer Methods and Advances in Geomechanics (IACMAG).
Abuiziah, I., Oulhaj, A., Sebari, K., Abassi Saber, A., Ouazar, D. & Shakameh, N. (2013). Modeling
and controlling flow transient in pipeline systems: Applied for reservoir and pump systems combined
with simple surge tank. Revue Marocaine desSciences Agronomiques et Vétérinaires, 1(3),12–18.
Amara, L., Berreksi, A. & Achour, B. (2013). Adapted MacCormack finite-differences scheme for water
hammer simulation.Journal of Civil Engineering and Science, 2(4), 226–233.
Bergant, A., Simpson, A.R. & Tijsseling, A.S. (2006). Water hammer with column separation: A histori-
cal review. Journal of Fluids and Structures, 22(2), 135–171.
Borga, A., Ramos, H., Covas, D., Dudlik, A. & Neuhaus, T. (2004). Dynamic effects of transient flows
with cavitation in pipe systems. The Practical Application of Surge Analysis for Design and Operation–
9thInternational Conference on Pressure Surges (Vol. 2, pp. 605–617). Bedfordshire, UK: BHR Group
Limited.
Brunone, B., Karney, B.W., Mecarelli, M. & Ferrante, M. (2000). Velocity profiles and unsteady pipe
friction in transient flow. Journal of Water Resources Planning and Management, 126(4), 236–244.
Chaudhry, M.H. & Hussaini, M.Y. (1985). Second-order accurate explicit finite-difference schemes for
water hammer analysis. Journal of Fluids Engineering, 107(4), 523–529.
106

Covas, D., Stoianov, I., Mano, J.F., Ramos, H., Graham, N. & Maksimovic, C. (2005). The dynamic
effect of pipe-wall viscoelasticity in hydraulic transients Part II—Model development, calibration
and verification. Journal of Hydraulic Research, 43(1), 56–70.
Covas, D., Stoianov, I., Ramos, H., Graham, N. & Maksimovic, C. (2004). The dynamic effect of pipe-
wall viscoelasticity in hydraulic transients Part I—Experimental analysis and creep characteriza-
tion. Journal of Hydraulic Research, 42(5), 517–532.
Etlender, R., Iben, U. & Bischoff, M. (2007). Simulation of wave propagation in flexible hoses. Fluid
structure interaction and moving boundary problems IV (pp. 131–141).
Gerbi, C.B.S. (2002). An implicit finite volume scheme for un steady flows in deformable pipe-lines,
Finite volumes for complex applications III (p. 461). France, Elsevier Science & Technology.
Greyvenstein, G. P. (2002). An implicit method for the analysis of transient flows in pipe networks.
International Journal for Numerical Methods in Engineering, 53(5), 1127–1143.
Hadj-Taıeb, E. & Lili, T. (2000). Validation of hyperbolic model for water-hammer in deformable
pipes. Journal of Fluids Engineering, 122(1), 57–64.
Heinsbroek, A.G.T.J. (1997). Fluid-structure interaction in non-rigid pipeline systems. Nuclear Engi-
neering and Design, 172(1), 123–135.
Heinsbroek, A.G.T.J. & Tijsseling, A.S. (1994). The influence of support rigidity on water hammer
pressures and pipe stresses. Proceedings of 2nd International Conference on Water Pipeline Systems
(pp. 17–30). Scotland: BHR Group.
Henclik, S. (2010). Mathematical model and numerical computations of transient pipe flows with fluid-
structure interaction. Transactions of the Institute of Fluid-Flow Machinery, 77–94.
Kaliatka, A. & Vaisnoras, M. (2005). Simulation of water hammer experiments using RELAP5 code
Proceedings of the International Conference on Nuclear Energy for New Europe, Bled, Slovenia
(pp. 541–549).
Keramat, A., Tijsseling, A.S., Hou, Q. & Ahmadi, A. (2012). Fluid-structure interaction with pipe-wall
viscoelasticity during water hammer. Journal of Fluids and Structures, 28, 434–455.
Lavooij, C.S.W. & Tusseling, A.S. (1991). Fluid-structure interaction in liquid-filled piping systems.
Journal of Fluids and Structures, 5(5), 573–595.
Li, Q.S., Yang, K. & Zhang, L. (2003). Analytical solution for fluid-structure interaction in liquid-
filled pipes subjected to impact-induced water hammer. Journal of Engineering Mechanics, 129(12),
1408–1417.
Mansuri, B., Salmasi, F. & Behrooz, O. (2014). Effects of pipe’s roughness and reservoir head levels on
pressure waves in water hammer. Journal of Civil Engineering and Urbanism, 4(1), 36–40.
Mitosek, M. & Szymkiewicz, R. (2012). Wave damping and smoothing in the unsteady pipe flow. Jour-
nal of Hydraulic Engineering, 138(7), 619–628.
Monajitha A.S., Sumam K.S. & Sajikumar N. (2014). Modelling of energy dissipation during transient
flow. International Journal of Scientific & Engineering Research, 5(7), 368–374.
Shen W., Zhang J. & Wang J. (2006), “Pipe Entrance Flow Simulation using Multigrid Finite Volume”,
Technical report, TR 460-06, Department of Computer science, University of Kentucky, Lexington, 2006.
Shu, J.J. (2003). Modelling vaporous cavitation on fluid transients. International Journal of Pressure
Vessels and Piping, 80(3), 187–195.
Simpson, A.R. & Bergant, A. (1994). Developments in pipeline column separation experimenta-
tion. Journal of Hydraulic Research, 32(2), 183–194.
Simpson, A.R. & Bergant, A. (1994). Numerical comparison of pipe-column-separation models. Jour-
nal of Hydraulic Engineering, 120(3), 361–377.
Singhal, A.K., Athavale, M.M., Li, H., & Jiang, Y. (2002). Mathematical basis and validation of the full
cavitation model. Journal of Fluids Engineering, 124(3), 617–624.
Soares, A.K., Covas, D.I., Ramos, H.M. & Reis, L.F.R. (2009). Unsteady flow with cavitation in viscoe-
lastic pipes. International Journal of Fluid Machinery and Systems, 2(4), 269–277.
Soares, A.K., Covas, D.I. & Reis, L.F. (2008). Analysis of PVC pipe-wall viscoelasticity during water
hammer. Journal of Hydraulic Engineering, 134(9), 1389–1394.
Streeter, V.L. & Wylie E.B. (1967). Hydraulic transients. New York, New York: McGraw-Hill.
Sumam, K.S., Thampi, S.G. & Sajikumar, N. (2010). An alternate approach for modelling of transient
vaporous cavitation. International Journal for Numerical Methods in Fluids, 63(5), 564–583.
Tijsseling, A.S. (1996). Fluid-structure interaction in liquid-filled pipe systems: A review. Journal of
Fluids and Structures, 10(2), 109–146.
Tijsseling, A.S. (1997). Poisson-coupling beat in extended water hammer theory (Vol. 53, pp. 529–532)
Dallas: ASME-Publications-AD.
Tijsseling, A.S. (2007). Water hammer with fluid-structure interaction in thick-walled pipes. Computers &
Structures, 85(11), 844–851.Tijsseling, A.S. & Heinsbroek, A.G.T.J. (1999). The influence of bend
107

motion on water hammer pressures and pipe stresses. Proceedings of 3rd ASME & JSME Joint Flu-
ids Engineering Conference, Symposium S-290 Water hammer (Vol. 248, pp. 1–7). ASME-FED, San
Francisco.
Tijsseling, A.S., Lambert, M.F., Simpson, A.R., Stephens, M.L., Vítkovský, J.P. & Bergant, A. (2008).
Skalak’s extended theory of water hammer. Journal of Sound and Vibration, 310(3), 718–728.
Tijsseling, A.S. & Vardy, A.E. (2004). Time scales and FSI in unsteady liquid-filled pipe flow. Proceed-
ings of the 9th International Conference on Pressure Surges (pp. 135–150). Chester, UK:BHR Group.
Tijsseling, A.S., & Vardy, A.E. (2005a). Fluid-structure interaction and transient cavitation tests in a
T-piece pipe. Journal of Fluids and Structures, 20(6), 753–762.
Tijsseling, A. & Vardy, A.E. (2005b). 20 years of FSI experiments in Dundee. Proceedings of the Third
MIT Conference on Computational Fluid and Solid Mechanics, Elsevier, Boston, USA. 2005.
Urbanowicz, K. & Zarzycki, Z. (2008). Transient cavitating pipe flow: Computation models and methods.
Task Quarterly, 12(3), 159–172.
Vardy, A.E. & Fan, D. (1989, October). Flexural waves in a closed tube Proceedings of the 6th Interna-
tional Conference on Pressure Surges (pp. 43–57). Cambridge, UK: BHRA.
Wiggert, D.C., Hatfield, F.J. & Stuckenbruck, S. (1987). Analysis of liquid and structural transients in
piping by the method of characteristics. Journal of Fluids Engineering, 109(2), 161–165.
Wiggert, D.C. & Tijsseling, A.S. (2001). Fluid transients and fluid-structure interaction in flexible liq-
uid-filled piping. Applied Mechanics Reviews, 54(5), 455–481.
Wilcox, J. (2012). Evaluating dynamic loads in piping systems caused by water hammer. Applied Flow
Technology.
108

Modified roughness equations for erodible bed open channel flow
L. Dhanesh & K.P. Indulekha

Department of Civil Engineering, College of Engineering Trivandrum, Kerala, India
ABSTRACT Study investigates the variation of the roughness values with flow depth,
channel slope and geometry. Four types of experimental setup were prepared in the labo-
ratory for the study, straight and curved channels with straight and trapezoidal cross sec-
tions. Variation of Mannings n and Chezys C with geometrical features and flow depth were
studied. It is found that the roughness values varies significantly in case of curved channels
and therefore prediction of discharge using this conventional roughness constants created-
significant error in discharge estimation. Dimensional analysis is carried out to incorporate
certain geometrical and flow parameters to the roughness equation. Roughness equations
(Mannings n and Chezys C) were modified and validated with the stage discharge data of
Vamanapuram river. It is found that modified equation gives more accurate discharge values
in erodible bed channel flows.
1 INTRODUCTION
Usual practice in one dimensional flow analysis is to select an appropriate value of rough-
ness coefficients for evaluating the discharge in open channel flow. This value of roughness
is often taken as uniform for entire cross section and entire surface and for all flow depths.
Researchers have studied the phenomenon of flow mostly in straight channels and have pro-
posed a number of methods to quantify the discharge. However these methods give better
results for straight channels but not for curved channels, because of formation of secondary
vortex flows in curve bends.
Under steady and uniform flow conditions, the equations proposed by Chezy or Manning
(1891) are used to compute the mean velocity of a channel.
Manning’s equation V = (1/ n )R1/2 S (1)

Chezy’s equation V = CR S 1/2 1/2
(2)
where, R = the hydraulic mean radius of the channel section, n = Manning’s roughness coef-
ficient, and C = Chezy’s channel coefficient, S = bed slope
Jarrett (1984) developed a model to determine Manning’s n for natural high gradient chan-
nels having stable bed and in bank flow without meandering coefficient. He proposed, a value
for Manning’s n as
n = ( 0.32 S 0.39 ) / R 0.16 (3)
This equation suggested by Jarret can be used for a channel with hydraulic radii from
0.15–2.1 m and gradient ranging from 0.002–0.04. Al-Romaih et al. (1999) reported the effect
of bed slope and sinuosity on discharge of two stage meandering channel. Based on dimen-
sional analysis, an equation for the conveyance capacity was derived
A series of experiments were conducted on compound channel in straight reaches under
isolated and interacting conditions by Pang (1998). He found that the Mannings n value var-
109

ied with flow depth. The value of n with the same surface in the main channel and floodplain
possessed different values when the water depth in the section varied
A new velocity formula, in the form of a modified Manning equation, is proposed for flow
prediction in alluvial channels by Guoliangyu and Lim S.Y. (2003). The new velocity formula
is used in conjunction with these resistance laws and it incorporates directly the bed geometry
on flow resistance.
In the present work an attempt is made to study the variation of roughness coefficients with
discharge, channel geometry and sinuosity under erodible bed condition. The effect of aspect
ratio, bed slope, roughness coefficients, depth of flow and sinuosity on discharge is analyzed.
Dimensional analysis is carried out to predict a new modified roughness coefficient.
2 METHODOLOGY
2.1 Experimental setup

The experimental setups were prepared in the laboratory. Mainly four types of flumes were
constructed to analyze the variation of roughness coefficients in open channel flow. Type I and
II is straight channel with rectangular and trapezoidal cross-section respectively whereas Type
III and IV is curved channel with rectangular and trapezoidal cross-section respectively.
2.2 Experimental procedure

Water is supplied to the channel by a recirculating system of water supply controlled by
pumps and gates. Plan view and cross section of the experimental setup is shown in the Fig-
ures 1, 2, 3 and 4. A layer of coarse aggregate is provided at the upstream of the channel to
reduce turbulence of the incoming water. Gates were provided both at upstream and down-
stream of the channel to maintain a uniform flow through the channel.
Water flows through the channel is collected in a masonry volumetric tank from where it is
allowed to flow back to the underground sump. From the sump, water is pumped back to the
overhead tank, thus setting a complete re-circulating system of water supply for the experi-
mental channel. The experiments were conducted under erodible bed condition. A thin layer
of about 10 cm thick cohesionless river sand was provided as bed material.
Point velocities were measured using area velocity meter at number of locations across
the predefined channel sections. The pitot tube was used to take velocities at lower depths.
Discharge in the channel was measured by the time rise method. The change in the depth of
Figure 1. Plan and cross sectionalview of Type I channel.
Figure 2. Plan and cross-sectional view of Type II channel.
110

Figure 3. Plan and cross sectional view of Type III channel.
Figure 4. Plan and cross sectional view of Type IV channel.
water with time was measured by stopwatch. A hand-operated tailgate weir was constructed
at the downstream end of the channel to regulate and maintain the desired depth of flow in
the channel. Readings were taken for the different slopes, discharges and aspect ratio. Around
44 readings were taken for the analysis from the four channel types.
3 RESULT AND DISCUSSIONS
3.1 Bed material and properties

Sieve analysis was carried on the sediment bed material and the particle size distribution
curve obtained is shown in the Figure 5. Since the uniformity coefficient is not greater than 6
and coefficient of curvature not within 1 and 3 the used river sand for experimental run was
poorly graded. The fineness modulus of the sand = 4.63, uniformity Coefficient Cu = 1.84,
and coefficient of curvature Cc = 0.81.
3.2 Experimental run

The experimental runs were carried out in all four channel type by varying depth. The roughness
coefficients Manning’s n and Chezys C were computed for each depth of flow and tabulated as
follows in Tables 1 and 2. Aspect ratio is the ratio of width of channel to the depth of flow.
3.3 Variation of Manning’s n with aspect ratio

The variation of Manning’s n with depth of flow for all the channel types is shown in Fig-
ure 6. Variation of Manning’s n with flow depth for Type I channel is found to be almost
uniform. Hence this equation can be used for discharge estimation in straight channel with
rectangular cross section.
It is found that the variation of Mannings n is too small or negligible for straight chan-
nels compared to curved channel. Slight variation in n is found at low flow depths. There is a
drastic variation in n value in curved channel. There may be heavy energy loss in bends, which
is seen as variation in n values. This may be due to the formation of secondary vortex flow in
bend portions of curved channels.
111

Figure 5. The particle size distribution curve.
Table 1. Details of various parameters in straight channel (TYPE I and TYPE II).
Channel Aspect Channel Aspect

type ratio Q(m3/s) V(m/s) R(m) n C type ratio Q(m3/s) V(m/s) R(m) n C
I 13.33 0.010 0.398 0.039 0.0126 46.2 II 15.00 0.003 0.440 0.0357 0.0107 53.41
9.09 0.019 0.494 0.054 0.0126 48.8 8.82 0.018 0.560 0.0570 0.0115 53.82
7.06 0.028 0.566 0.066 0.0125 50.5 8.00 0.023 0.575 0.0619 0.0119 53.00
6.00 0.036 0.615 0.075 0.0125 51.5 7.59 0.028 0.590 0.0647 0.0119 53.20
5.00 0.048 0.673 0.087 0.0125 52.7 6.52 0.033 0.630 0.0735 0.0121 53.29
Table 2. Details of various parameters in curved channel (TYPE III and TYPE IV).
Channel Aspect Channel Aspect

type ratio Q(m3/s) V(m/s) R(m) n C type ratio Q(m3/s) V(m/s) R(m) n C
III 16.00 0.005 0.480 0.032 0.0092 61.55 IV 20.00 0.0045 0.338 0.022 0.010 51.36
12.80 0.007 0.500 0.038 0.0099 58.77 15.15 0.0065 0.365 0.029 0.011 48.84
11.64 0.008 0.500 0.040 0.0104 56.69 11.11 0.0099 0.395 0.038 0.013 46.02
10.67 0.009 0.504 0.043 0.0107 55.35 10.00 0.0115 0.409 0.042 0.013 45.48
9.85 0.010 0.498 0.045 0.0112 53.43 7.69 0.0165 0.436 0.053 0.014 43.37
Figure 6. Variation of Manning’s n with aspect ratio.
3.4 Variation of Chezys C with aspect ratio

The variation of Chezys C with flow depth is investigated for straight and curved channel and
is shown in Figure 7. It can be seen from the figure that curved channels (Type III and Type IV)
exhibits a steady increase in the value of C with aspect ratio, indicating that the curved channel
consumes more energy as the depth of flow increases. For straight channels (Type I) the variation
is small compared to curved channel. In curved channels especially at bends velocity predomi-
nates in all the three directions where as in straight channel velocity is moreover unidirectional.
Variation of the roughness values were evaluated in terms of standard deviation and
presented in Table 3. Chezys C has higher standard deviation values (13.12) compared to
Manning’s n (0.0018). Higher value of standard deviation indicates lesser accuracy in com-
putation of discharge. Hence Manning’s n can be used to predict discharges in open channel
112

Figure 7. Variation of Chezys C with aspect ratio.
Table 3. Standard deviation of Mannings n and Chezys C.
SL Roughness Standard
no Coefficients Mean deviation
1 Mannings n 0.013 0.0018

2 Chezys C 41.94 13.12
with more accuracy than Chezys C. Variation of these roughness coefficients points towards
the necessity of modifying these equations.
3.5 Dimensional analysis

The parameters taken for the dimensional analysis are hydraulic radius R, velocity of flow U,
gravitational acceleration g, kinematic viscosity ν, bed slope S, aspect ratio ∝ and sinuosity
Sr, and it can be expressed functionally as,
Roughness coefficients = ϕ (R, U, g, ν, S, Sr, ∝)
Conventional Mannings equation is
1 2 /3
U= R S (4)
n
Multiplying equation (4) with [ Rv ] to get,

1
 10 1  2
UR 1  R 3 S 2 
=  2  (5)
v n v 
 
Again since, conventional Chezys equation given by the equation (6)

U = CR1/ 2S1/ 2 (6)
Multiplying equation (6) with [ Rv ] to get

UR C 2 /3 1/ 2
= R S (7)
v v
The sinuosity ratio is inversely proportional to velocity, slope and aspect ratio. Hence the
eqn. (5) may be rewritten in the following form as
  103  
UR 1 gR 
= ϕ  , S , ∝,   (8)
v  Sr  v2m 3  
1
   
113

As manning’s roughness coefficient is dimensionally non-homogenous, a length factor m1/3
is considered to it to make it homogenous. After dimensional analysis, relation obtained as
 1

URSr  gR10 /3s 2 
vs  1 
(9)
v∝  v 2 m 3 
Graph is plotted between these dimensionless parameters to obtain a simple relationship

between these parameters as shown in Figure 8. Power equation with regression coefficient
value of 0.9481 has given better relation for these parameters
With this power equation Mannings constant n is modeled as follows
0.7616
 10 /3 12 

URSr  gR s  (10)
= 0.0065  1 
v∝  v 2 m 3 
The newly developed relationship for Manning’s n constantin open channel flow is
Sr v 0.52 gs 0.1192 m 0.2538

n= (11)
0.0065 ∝ R 0.878 g 0.7616
Similarly, eqn. 9 can also be rewritten for Chezy’s C using dimensional

3
analysis. Chezy’s
C is dimensionally non-homogenous. So, the dimensional group ( gR v 2 ) is used because of its
similarity to the traditional Chezy’s equation for one dimensional flow in a prismatic chan-
nel. Both give similar equation which is expressed as
UR 1  gR3  
= ϕ  , S , ∝,  2   (12)
v  Sr  v 
Then graph is1plotted between these dimensionless parameters

3
URSr
v∝ and gR 2S 2 to obtaina simple relation between these parameters shown in Figure 9.
v
After plotting between dimensionless parameters, the best possible combination is found
to be given by the power equation having a regression correlation of 0.939 as
0.8421
 3 12 

URSr  gR s  (13)
= 0.0007  2 
v∝  v 
The newly developed relationship for Chezys constant C in open channel is expressed as
0.0007 g 0.8421R1.04 ∝
C= (14)
s 0.079v 0.684Sr
10
URSr
Figure 8. Calibration equation between ( gR 3
) and v∝ for Mannings n.
v2 m 1
3
114

3.6 Comparison of results
The values for Mannings n and Chezys C are calculated from the newly modeled equations.
These values are used to compute discharge (Modeled). Modeled equation gave more accu-
rate values of discharge for both in straight and curved channels. Variation of modeled dis-
charge from observed discharge for all channel types are plotted in Figures 10 and 11.
It is clear from the figure that both in case of Chezys C and Mannings n the newly modified
equation gives better results that is close to the actual observed discharges, as it incorporated
geometric features of channel and flow parameters. Modified equation of Mannings n gave
better results especially at lower discharges, as energy loss is much less in low discharges. Fig-
ure 12 shows the correlation values of observed and modeled discharge. Mannings equation
gives correlation coefficients of 0.9804 and Chezys equation gives correlation coefficients of
0.9713. Hence Mannings n can give better results than Chezys C.
3.7 Validation of result

The obtained modified equation for Manning’s n and Chezys C is validated with the actual stage
discharge of Vamanapuram river. Actual stage-discharge and cross-sectional details at Ayilam
station is obtained. Discharge values were computed with modified roughness equation. It is
found that computed discharge using modified equation lays more proximity to actual value.
Around 210 no of data were analyzed for the validation and graphically presented in Figure 13.
10
3
Figure 9. Calibration equation between ( gR ) and URSr
v∝
for Chezys C.
v2 m 1
3
Figure 10. Variation of observed and modeled discharge using Mannings n for all channel types.
Figure 11. Variation of observed and modeled discharge using Chezys C for all channel types.
115

Figure 12. Calibration for Mannings n and Chezys C.
Figure 13. Validation of proposed Mannings n.
4 CONCLUSIONS
Based on the analysis and discussions following conclusion were drawn.

• Roughness coefficients n and C changes with aspect ratio especially in curved channels due
to occurrence of secondary vortex flow at bend apex of curved channels
• Dimensional analysis is carried out and it is found that the roughness constants depend on four
dimensionless parameters Reynolds number, sinuosity, longitudinal slope and aspect ratio
• A modified equation to find roughness coefficients is formulated and the proposed equa-
tion is found to give better result compared to conventional equations for a movable bed
of poorly graded river sand with uniformity coefficients less than 2 and coefficient of
curvature less than 1
ACKNOWLEDGMENT
The authors gratefully acknowledge the contributions from the department of Civil Engi-
neering, College of Engineering Trivandrum in providing the necessary infrastructure for
carrying out the study.
REFERENCES
Al-Romaih, Shiono K. and Knight D.W. (1999). “Stage Discharge Assessment in Compound Meander-
ing Channels”, Journal of Hydraulic Engineering. ASCE, vol.125, No. 1, pp. 66–77.
Guoliang Yu and Lim S.Y. (2003). “Modified Manning formula for flow in alluvial channels with sand-
beds”, Journal of hydraulic research, Volume 41, 2003 – issue 6, pp. 597–608.
Jarrett R.D. (1984). ‘‘Hydraulics of high gradient streams’’, Journal of Hydraulics Engineering, ASCE,
110, pp. 1519–1539.
Pang B. (1998). “River flood flow and its energy loss”, J. Hydr.Engrg., ASCE, 124(2), 228–231.
116

Economic design of a surge tank: An alternative approach
Surendran Thara, K.S. Sumam & N. Sajikumar

Department of Civil Engineering, Government Engineering College Thrissur, Thrissur, India
ABSTRACT: Surge tanks are designed to protect a hydropower plant against the pressure
surges caused by sudden load changes. The design parameters of a surge tank are significantly
influenced by the pressure surges developed during valve operations. In this study, the pressure
variations during valve closure are taken to calculate the corresponding water level fluctuations
in the surge tank using computational fluid dynamics. These fluctuations are again computed
by incorporating fluid–structure interaction, which accounts for the damping of pressure surges
caused by the vibration of piping components. On comparing the results obtained from numeri-
cal analysis with and without considering fluid–structure interaction, it was found that the water
level fluctuations are suppressed and the system stabilizes quickly, considering the effect of fluid–
structure interaction. Hence, the size of the surge tank, which directly depends on the magnitude
of pressure surges during transient flow, can be reduced by considering the effect of fluid–struc-
ture interaction. Therefore, this paper presents an economic design of a surge tank by reducing its
diameter due to the incorporation of fluid–structure interaction in the design.
Keywords: fluid–structure interaction, computational fluid dynamics, pipe flow, surge con-
trol, transient analysis
1 INTRODUCTION
A water hammer is defined as a pressure surge caused when fluid motion is forced to stop sud-
denly. Due to the effect caused by water hammer, a pressure wave is created which travels from
the point of generation through the piping system. This motion will disturb the entire piping
system. The intensity of this pressure wave is such that it sometimes leads to the complete
failure of the system. There are many methods that can be used to control the undesirable
transients in pipeline systems and to reduce their negative effects. One such popular method,
used in hydropower plants, is the installation of a surge tank. The purpose of the surge tank is
to intercept and dampen these high pressure waves and not allow them to drop toward vapor
pressure. In the present study, the damping effect of a surge tank is studied and analyzed.
A large amount of research has been conducted to study the damping behavior in surge tanks.
The height of a surge tank is decided based on the highest possible water level during the transient
state, such as at the time of sudden valve closure. Traditional methods for deciding the size of a
surge tank are based on the magnitude of the pressure surge obtained from transient analysis.
Research by Moghaddam (2004) provided analytical equations for the analysis and design of a
simple surge tank. Equations for maximum and minimum surge and its time of occurance were
also determined. Another set of numerical equations were developed by Nabi et al. (2011) and
Abuiziah et al. (2014) using the Runge-Kutta method. However, using such equations requires con-
siderable computational time and effort. Other methods developed include a user friendly chart by
Seck and Fuamba (2015) and software developed for modeling the transient phenomena in a surge
tank. One of the popular software products is computational fluid dynamics. Recently, reserachers
like Lan-Lan et al. (2013) and Torlak et al. (2015) used computaional fluid dynamics for modeling
transient phenomena. Both considered two phase flow in the surge tank and used a volume of fluid
approach and K-ε turbulance model. From the study by Torlak et al. (2015), it was found that the
result from computational fluid dynamics is close to the real scale situation. Hence, in the present
117

study, the transient analysis is carried out using computational fluid dynamics by a volume of fluid
approach and K-ε turbulance model. Many studies successfully modeled the transient phenomena
of surge tanks but did not consider the damping effects caused by the structural interaction. From
the literature, it is clear that the volume of fluid approach method is advantageous due to its free
surface capturing property and the K-ε model can be useful in the turbulant conditions occurring
in surge tanks during transient flow and therefore, are used in the present study.
The idea about fluid–structure interaction was observed in the research conducted by Hou
et al. (2012). He mentioned the importance of fluid–structure interaction and its importance
in structural design.
This paper attempts to find the results required in the actual scenario for the surge tank
design. A comparison is done between the fluid model and fluid–structure interaction during
transient conditions. Various stages of the study are:
• Transient analysis of the surge tank using Computational Fluid Dynamics (CFD)
• Transient analysis of the surge tank using CFD by incorporating Fluid Structure Interac-
tion (FSI)
• Comparison of above results
• Analyze the surge tank with reducing it’s size (diameter) and repeating the above process
to check the achievement of desired performance.
2 DETAILS OF STUDY AREA
The study area selected for the transient analysis is Pallivasal hydroelectric power plant which
is located in Munnar in the Idukki District. The Munnar region is situated at an altitude of
1500–2500 m above sea level and has an average rainfall of 275 cm. The hydraulic system
consists of a head race tunnel of 3396 m long and a pressure shaft of 2500 mm diameter. The
head race tunnel and pressure shaft are common to both powerhouses. The pressure shaft is
bifurcated into two penstocks of 2000 mm and 1600 mm diameter to feed the extension and
existing power houses. These penstocks are further divided in such a way that it has two num-
bers of 1500 mm diameter for the extension scheme and three numbers of 460 mm diameter
and three numbers of 583 mm diameter for the existing generating scheme. Figure 1 shows
the layout of Pallivasal hydroelectric power plant.
3 COMPUTATIONAL MODEL
The geometric model includes a tank, that is a surge tank having a 7 m diameter and two
pipes; the head race tunnel having a 3.5 m diameter and the penstock having a 2.5 m diameter.
Figure 1. Pallivasal water conductor system.
Figure 2. Geometry of surge tank.
118

Both of these are named as inlet and outlet, see Figure 2. Due to the difficulty in the modeling
of a large network, only a small portion of the geometry is modeled by giving appropriate
boundary conditions.
3.1 Initial conditions

The initial conditions are provided by the steady state analysis. For the steady state condi-
tions, boundary conditions are given as:
• Inlet pressure: 184722.3 Pa
• Top of surge tank: 0 Pa
• Outlet: Mass flow rate: 22599.99 Kg/s
The initial conditions are common for both fluid and FSI model.
4 TRANSIENT ANALYSIS FOR THE FLUID MODEL CONDITION
After attaining a steady state flow condition, the transient condition generated in the surge
tank is taken from the work of Shima (2013) and shown in Figure 3 and Figure 4.
The pressure values in Figure 3 and Figure 4 were extracted using Grabit software in Mat-
lab. Due to the modeling difficulty of the entire geometry of the hydraulic system in Pallivasal
hydropower plant in Fluent, the boundary condition applied at the outlet during transient
analysis is slightly modified based on the target. The values given in Figure 3 and Figure 4 are
the pressure values at the turbine ends of two units. Since the entire piping system is not mod-
elled for getting the boundary conditions, it is necessary to reduce these values while applying
on the given geometries outlet. Initially, the transient analysis was carried out for the fluid
model. The obtained water level oscillations in the pressure surges at Unit 1 and at Unit 6 are
given in Figure 5 and Figure 6. It was found that the water level slightly dampened toward the
end of simulation.
i. Surge oscillations in the surge tank due to load rejection at Unit 1
From Figure 5, the maximum water level during analysis was found to be 42 m and the mini-
mum water level was 15 m. The oscillation continues with a reduction in water level for more
than three seconds.
ii. Surge oscillations in the surge tank due to load rejection at Unit 6
From Figure 6, the maximum water level during transient condition was observed as
43 m and the minimum water level was 14 m. The oscillation continued for more than
two seconds.
Figure 3. Pressure head at existing Unit 1

Figure 4. Pressure head at existing Unit 6
while providing 7 m diameter surge tanks.
while providing 7 m diameter surge tanks.
Figure 5. Surge oscillations in Unit 1. Figure 6. Surge oscillations in Unit 6.
119

5 TRANSIENT ANALYSIS WITH FSI
Transient analysis was again carried out for the same surge tank, for the same condition
incorporating the effect of FSI, which considers the coupling mechanism between the fluid
and the structure using the principles of system coupling. The simulated water levels at the
surge tank for Units 1 and 6 are represented in Figure 7 and Figure 8.
Figure 7 shows a graph between head and time for the surge tank at the time of transient
analysis for Unit 1. It was observed that the maximum water level was 38 m and the minimum
water level was 14 m. The water level dampened completely at 0.9 seconds.
Figure 8 shows the water level variation in the surge tank while considering the transient
condition at Unit 6 with the incorporation of fluid–structure interaction. It was found that
the maximum water level was 38 m and the minimum water level was 15 m. The pressure wave
dampened completely at 0.8 seconds.
6 COMPARISON OF RESULTS
After completing the fluid and FSI analyses for the two units during the transient condi-
tion, the next step was to compare and analyze the results. Table 1 shows the variation of
results in both units.
The water level fluctuations obtained from the FSI model were then compared with the fluid
model and the results are depicted in Figure 9 and Figure 10. It was found that the FSI model
had less peak compared to the fluid model. However, the lag between the curves increases dras-
tically as the wave progresses. It can also be seen that the FSI model is subject to more damp-
ing compared to the fluid model. FSI predicts surge tank oscillations with additional energy
Figure 7. Water level variation in the surge tank Figure 8. Water level variation in surge tank
considering FSI (Unit 1). considering FSI (Unit 6).
Table 1. Comparison of water level of oscillation in the surge tank with and without FSI.
Analysis
With FSI Without FSI
Surge tank diameter Time Maximum Minimum Time Maximum Minimum
Unit 1 0.9 s 38 m 14 m 3s 42 m 15 m
Unit 6 0.8 s 38 m 15 m 2s 43 m 14 m
Figure 9. Comparison of head result based on Figure 10. Comparison of head result based on
Unit 1. Unit 6.
120

Figure 11. Head variation in the surge tank (fluid Figure 12. Head variation in surge tank (FSI
only). case).
dissipation. Hence, the design size of the surge tank can be reduced, because in earlier case
design considered the effect of fluid only. For a given load change, the surge amplitude in a
simple chamber is approximately proportional to the diameter of the chamber, if the cham-
ber is big enough, the surge becomes “deadbeat” and will die away after the half cycle. A simi-
lar deadbeat condition will result if the load change is sufficiently slow. Deadbeat chambers
are not usually economical (Nabi et al., 2011). Hence, the design size of the surge tank can be
reduced, as selected earlier by transient analysis (fluid model).
7 DIAMETER CHANGE UNDER ECONOMIC CONSIDERATION
From the above result one can say that the FSI dampens rapidly and the head developed in
the surge tank with FSI is less compared to the fluid model. Therefore, in order to avoid the
deadbeat condition in a surge tank for a large magnitude pressure wave, it is necessary to find
the performance of a surge tank based on a reduced diameter. Hence, a surge tank of reduced
diameter (i.e. 6 m) was selected and the procedure repeated as above
7.1 Steady state analysis

Steady state analysis for the 6 m diameter surge tank was initially conducted as per section 3.1.
7.2 Transient analysis for reduced size

For simulating the sudden closure of the valve, the boundary condition at the outlet is giving
from the Figure 4. From the transient analysis water level in the surge tank were obtained.
Water level oscillations in the surge tank for Unit 6 are given in Figure 11.
From Figure 11, the maximum water level obtained was 44 m and the minimum water level
was 14 m.
7.3 Fluid structure interaction

Transient analysis was again carried out for the same surge tank, for the same unit incorpo-
rating the effect of FSI, which considers the coupling mechanism between the fluid and the
structure using the principles of system coupling. The water level oscillations at the surge
tank are plotted in Figure 12 for Unit 6. The maximum head in the surge tank was found to
be 42.5 m and the minimum head was found to be 14 m, with the wave dampening completely
at 0.6 seconds. Hence, it was found that the reduction in size of the surge tank did not com-
promise the performance of the surge tank with a 7 m diameter.
8 COMPARISON OF RESULT
Comparing the water levels in the surge tank with the change in size, it was observed that by
reducing the diameter, the maximum water level will go up to 42.5 m and the minimum water
level go down to 14 m. However, it is still a safe version.
While observing the result from the two analyses of reducing diameter, it was observed
that the FSI model gives a small water level fluctuation compared to the fluid model, which
developed a high head in the surge tank during the transient condition. The comparison of
water level fluctuation in the surge tank for fluid and FSI cases is given in Figure 13, which
clearly indicates the damping mechanism during fluid–structure interaction.
121

Table 2. Comparison of water levels for different diameter surge tank (Unit 6).
Water level
With FSI Without FSI
Surge tank diameter Time Maximum Minimum Time Maximum Minimum
7m 0.8 s 38 m 15 m 2s 43 m 14 m
6m 0.6 s 42.5 m 14 m 1s 44 m 14 m
Figure 13. Comparison of water level fluctuation in the surge tank in fluid and FSI cases.
9 CONCLUSIONS
Generally, the size of the surge tank is decided based on the magnitude of the pressure surge
obtained from transient analysis which does not consider the damping effect caused by the
fluid–structure interaction.
Based on this study, the following conclusions were made:
• The design size of the surge tank, which directly depends upon the magnitude of the pres-
sure surge, can be reduced without compromising performance.
• System stabilization occurs at an early stage when FSI is considered.
• The cost of construction of the surge tank can be minimized by obtaining the optimum
size of surge tank, hence design economy can be obtained.
REFERENCES
Abuiziah, I., Ouihaj, A., Sebari, K. & Ouazar, D. (2014). Comparative study on status and development
of transient flow analysis including simple surge tank. International Science Index, Civil and Environ-
mental Engineering, 8, 228–234.
Hou, G., Wang, J. & Layton, A. (2012). Numerical methods for fluid structure interaction—A review.
Communications in Computational Physics. 12(2), 337–377.
Lan-Lan., Zheng-Ghang., Jie., Dong & Guang-sheng. (2013). Numerical study of flow fluctuation
attenuation performance of a surge tank. Journal of Hydrodynamics, 6, 938–943.
Moghaddam, A. (2004). Analysis and design of asimple surge tank. International Journal of Engineering
Transactions A: Basics, 17, 339–345.
Nabi, G., Rehman, H., Kashif, M. & Tariq, M. (2011). Hydraulic transient analysis of surge tanks: Case
study of Satpara and Golen Gol hydro power projects in Pakistan. Pakistan Journal of Engineering
and Applied Sciences, 8, 34–48.
Ramdan, A & Mustafa, H. (2013). Surge tank design considerations for controlling water hammer
effects athydroelectric power plant. University Bulletin, 3, 147–160.
Sanchez, H., Cortes, C. & Matias, A. (2004). Structural behavior of liquid filled storage tanks of large
capacity placed in seismic zones of high risk in Mexico. 13th World Conference on Earthquake Engi-
neering, Vancouver B.C., Canada, 2665.
Seck, A. & Fuamba, M. (2015). Contribution to the analytical equation resolution using charts for
analysis and design of cylindrical and conical open surge tank. Journal of Water Resource and Pro-
tection, 7, 1242–1256.
Shima, S.M.S. (2013). Optimization of surge control in hydroelectric power plant. (Project report of
Govt. Engineering College, Thrissur).
Torlak, M., Bubalo, A. & Dzafervoic, E. (2015). Numerical analysis of water elevation in a hydropower
plant surge tank, International Association for Hydro-Environment Engineering and Research, 9, 9–11.
122

Degradation study on coir geotextiles treated with hydrocarbon

mixed soil
A.V. Praseeja & N. Sajikumar

Government Engineering College, Thrissur, India
ABSTRACT: This work aims to investigate the degradation characteristics of coir geo-
textiles in the presence of hydrocarbon, which is essential if the geotextiles are meant to be
used as a remedial measure against hydrocarbon contamination of soil. Several studies have
already reported on degradation of coir geotextiles for various civil engineering applications.
However, its degradation behavior in a hydrocarbon contaminated environment has not been
studied yet. In this study, experimental tests were conducted to explore the durability of coir
geotextiles with the effect of hydrocarbons. The experiment was carried out for one year to
investigate the time-dependent degradation behavior. The results show that the degradation
rate is within the normal range of values compared with previous studies. Hence, the degra-
dation rate of coir geotextiles is not much influenced by the action of hydrocarbons.
1 INTRODUCTION
Natural fibers are now a keen area of interest in many applications as they are recyclable and
readily available. Because of their availability and biodegradability, natural fibers can be used
to replace synthetic polymers in various applications. However, it is an inevitable fact that
the strength of natural fibers is continuously reduced on interaction with wet soil or water, or
sometimes under the action of micro-organisms or due to chemical reactions. This decaying
process is termed biodegradation, an important factor that needs to be considered before
these materials can be put to any engineering use (Joy et al., 2011).
Coir is a hard and tough organic fiber extracted from the husk of the coconut. Sri Lanka
and India are the major fiber producers in the world. About 55 billion coconuts are harvested
annually in the world, of which, only 15% of the husks are used for extracting fibers and the
rest abandoned in nature, hence becoming a cause of environmental pollution (Praveen et al.,
2008). Coir fiber is rich in cellulose (36 to 43% by weight) and lignin (41 to 45% by weight);
hence, it is regarded as an environmental friendly material. It also contains constituents such
as hemicellulose (0.15 to 0.25 wt%), protein (3 to 4 wt%) and ash (1 to 2 wt%) (Yao et al.,
2012). Due to its high lignin content, compared to other natural fibers such as jute, flax, linen
and cotton, coir is known to be the strongest of all natural fibers. From a study conducted
by Mukkulath and Thampi (2012), it is clear that the degradation of jute geotextile is faster
(within one year) than that of coir geotextile (eight to ten years). Coir fibers take 15 times
longer than cotton and seven times longer than jute to degrade (Rao & Balan, 2000). The
coir geotextile made of these coir fibers has already been proven as a material with excellent
qualities in various engineering applications. Most of the studies conducted in our country
show that coir geotextiles are preferable to other natural fibers for engineering applications
as they are readily available, cost-effective and environmentally friendly (Praveen et al., 2008;
Rao & Balan, 2000; Mukkulath & Thampi, 2012).
The advantages of natural lignocellulosic fibers also include acceptable specific strength
properties, low cost, low density and biodegradability (Rajan & Abraham, 2007). Praveen
et al. (2008) present a viable and cost-effective technology using coir geotextiles for the removal
of organic matter from wastewater. Mukkulath and Thampi (2012) have conducted a pilot
123

plant study by utilizing coir geotextile media of both woven and non-woven type as attach-
ment media for the biofilters for low volume organic-rich industrial wastewater. Non-woven
geotextile was found to be better for removing nitrate and phosphate from sample wastewater.
Schurholz (1991) conducted incubator tests to determine the reduction in strength of coir
geotextiles embedded in highly fertile soil. Coir geotextiles could retain 20% of their original
tensile strength after one year and showed no damage even after being kept wet for 167 days.
Rao and Balan (2000) conducted degradation studies using both retted and unretted coir
yarns in different soil media. The main observations were as follows: (i) The degradation rate
depends on the type of embedment soil, climatic conditions, organic content, moisture con-
tent and the type of coir used; (ii) The major loss of strength takes place in four to eight
months, which is more for sandy soil compared to that of clay; (iii) Brown coir degrades at a
faster rate than white coir; (iv) Degradation is more predominant at a pH of 6 and 8; and (v)
Degradation is faster in water than in soil. On observing the influence of embedment of coir
yarns in the natural climatic condition for one year in two different states (Rao & Balan, 2000),
it may be noted that: (i) The embedment of coir yarn in the soil of Kerala State has shown
a faster rate of degradation than in Delhi due to high organic and moisture content; and (ii)
The natural geotextiles, when kept at shallow depths, degrade at a faster rate. Joy et al. (2011)
conducted biodegradation studies on coir geotextile fibers in an experimental pit, where the
geotextiles were placed at different depths in the field. This study highlighted the complex-
ity of biodegradation studies, as they are influenced by various factors such as temperature,
humidity, type of yarn, pH, organic content, the moisture content of the soil and so on.
The efficiency of using natural fibers with high cellulose content in reducing the effect
of oil contamination over water bodies is discussed by Hubbe et al. (2013) and Karan et al.
(2011). These studies suggest coir geotextiles as environmentally friendly oil sorbents. While
using coir geotextiles as oil sorbents, it is necessary to examine their degradation behavior in
such an environment. Limited studies are available for meeting this end. Hence, in the present
study, experimental work was conducted to investigate the degradation of coir geotextiles in
a hydrocarbon mixed soil medium.
Table 1. Properties of materials used.
Sl no. Material type and property Value
1. Coir geotextile 900 GSM 600 GSM

i Tensile strength (kN/m) 1.95 1.208
ii Thickness (m) 0.012 0.006
2. Petrol
i Density (kg/m3) 746
ii Surface tension (N/m) 0.029
iii Kinematic viscosity (m2/s) 26e-06
3. Soil
i Type Silty sand
ii pH 6
Figure 1. Experiment setup.
124

In this study, non-woven coir geotextiles of 600 GSM and 900 GSM, size 20 cm x 20 cm were
used for the degradation study. This was supplied by the Coir Board, Thrissur, Kerala, India.
The degradation behavior of the samples was assessed by embedding them in wet lateritic soil
(pH 6) mixed with petrol (hydrocarbon) for 12 months. Table 1 shows the properties of the
materials used for the experiment.
The soil was collected from one meter below ground level and sun-dried for two days.
Lumps and debris were also removed. For setting the samples in one tray, 10 kg of clean soil
was collected, a small quantity of water was sprinkled over it and uniformly mixed with pet-
rol (0.5 liter). The evenly mixed soil was filled as two layers, and the coir samples were placed
in between the first and second layers. The whole tray was then covered with a plastic sheet
to avoid quick evaporation and holes were provided for aeration. Figure 1 shows the steps
involved in placing the samples. The rate of degradation was obtained by conducting tensile
strength tests on coir geotextile samples after treating them for one month, three months, six
months, nine months and 12 months.
3 TENSILE STRENGTH TEST
The tensile strength test is usually performed to determine the durability/degradation effect of
coir geotextiles, following the procedure outlined in IS 13162 (Part 5) and ASTM D 4595. The
tests were conducted at the Central Coir Research Institute (CCRI) in Alleppey, Kerala, India.
The test was conducted as per the procedure laid down in IS13162. The tensile properties of
the test specimens were obtained as tensile strength in kN/m, which denotes the maximum
resistance to deformation per unit width developed for a specific material when subjected to
tension by one external force. Three samples were tested for each trial, and the average value
was taken as the tensile strength of the specimen. The test setup is shown in Figure 2.
The tensile strength of coir geotextiles was calculated using the formula given in Equa-
tion 1 (IS: 13162 – Part 5).
af = Ff C (1)
Figure 2. Test setup for determining tensile strength of non-woven coir geotextiles.
125

where af = the tensile strength in kN/m of width; Ff = the observed maximum load (in kN);
and C = 1/B for non-wovens, where B = specimen width in meters.
The tensile strength behavior of non-woven CG (900 GSM and 600 GSM) were studied at
different time periods. The samples were embedded for one year to study the time-dependent
degradation behavior under the action of hydrocarbons. Tensile strength curves (Figure 3)
were obtained for both types of samples after treating them for one month, three months,
six months, nine months and 12 months, and the corresponding stress–strain behavior is
presented in Figure 4 and Figure 5.
The tensile strength of the geotextile is characterized by the observed maximum breaking
load. The strength corresponding to zero period indicates the strength of the untreated sam-
ple. The percentage strength reductions for each month are shown in Table 2.
In the period of three to six months, the percentage reductions in strength were observed
as about 52% and 72% for the 900 GSM and 600 GSM samples, respectively. In the case
of 900 GSM samples, a slight regain of strength was observed in the period of the sixth to
ninth month, but in month 12 it is reduced to 61.54%. In the case of 600 GSM samples, 90%
of strength reduction was observed in the ninth month, and a slight regain of about 18% in
strength noticed in the period of months 9–12.
The results obtained were compared with the degradation tests conducted by Rao and
Bala, (2000) on coir yarns embedded in different environments for one year. They observed
that the degradation rate of coir fibers mainly reflected the influence of pH of media and the
admixtures. Many studies highlight the significance of pH in the reduction of tensile strength
of coir geotextiles (Joy et al., 2011; Mukkulath & Thampi, 2012).
The pH value of the soil considered in this study is 6, which can cause around 66% of
degradation as per the results presented by Rao and Balan (2000). As the added substance,
in other words, hydrocarbon, was a non-polar compound (pH value is neutral), pH may not
have an influence on the results of this study. The observed overall strength reductions when
embedded in a hydrocarbon environment were about 61.54% for 900 GSM and 73.10% for
600 GSM samples, respectively. Hence, the degradation rate of geotextiles is not much influ-
enced by the action of hydrocarbons.
From the literature, it is also clear that cellulosic fibers possess the capacity for absorb-
ing oil content due to their high specific area (Hubbe et al., 2013; Karan et al., 2011). The
strength reduction in the case of geotextile samples pertain to these justifications. Payne et al.
(2012) reveal the ability of cellulosic fibers to absorb oil even in wet conditions. Within a
Figure 3. Tensile strength curve.
126

Figure 4. Stress–strain curve (600 GSM CG).
Figure 5. Stress–strain curve (900 GSM CG).
Table 2. Percentage tensile strength reduction.
900 GSM 600 GSM
Time Tensile strength % Strength Tensile strength % Strength

period (months) (kN/m) reduction (kN/m) reduction
Untreated 1.95 0 1.208 0

One 1.78 8.72 0.668 44.70
Three 0.93 52.31 0.345 71.44
Six 0.935 52.05 0.331 72.60
Nine 1.18 39.49 0.105 91.31
Twelve 0.75 61.54 0.325 73.10
period of one year, some portion of hydrocarbons may get absorbed into the coir fiber due to
its high specific area. Subsequently, the fibers become brittle which may lead to lower tensile
strength. Hence, absorption capacity studies and morphological analysis are beneficial for
identifying their suitability against hydrocarbon contamination.
127

5 CONCLUSION
This degradation study on non-woven coir geotextile media was intended to focus on the
application of natural fibers against hydrocarbon contamination. The results show that the
degradation rate under the action of hydrocarbon is almost same as the degradation rate
under other environmental conditions. Hence, the degradation rate of coir geotextiles is not
much influenced by the presence of hydrocarbons. The result also reveals the necessity of
further experimental studies for the proper understanding of the variation in their structure,
physical and chemical compositions. This can be considered as the future scope.
REFERENCES
ASTM D−4595. (2009). ASTM D4595–11 Standard Test Method for Tensile Properties of Geotextiles by
the Wide-Width Strip Method. West Conshohocken, USA: ASTM International.
Hubbe, M.A., Rojas, O.J., Fingas, M. & Gupta, B.S. (2013). Cellulosic substrates for removal of pollut-
ants from aqueous systems: A review. 3. Spilled oil and emulsified organic liquids. BioResources, 8(2),
3038–3097.
IS 13162: Part 5 (1992). Determination of tensile properties using a wide width strip. Bureau of Indian
Standards, New Delhi, India.
Joy, S., Balan, K. & Jayasree, P.K. (2011). Biodegradation of coir geotextile in tropical climatic condi-
tions. Proceedings of the Golden Jubilee Indian Geotechnical Conference, Kochi, India, 604–606.
Karan, C.P., Rengasamy, R.S. & Das, D. (2011). Oil spill cleanup by structured fibre assembly. Indian
Journal of Fibre & Textile Research, 36(2), 190–200.
Mukkulath, G. & Thampi, S.G. (2012). Performance of coir geotextiles as attached media in biofilters
for nutrient removal. International Journal of Environmental Sciences, 3(2), 784–794.
Payne, K.C., Jackson, C.D., Aizpurua, C.E., Rojas, O.J. & Hubbe, M.A. (2012). Oil spills abatement:
Factors affecting oil uptake by cellulosic fibers. Environmental Science and Technology, 46(14),
7725–7730.
Praveen, A., Sreelakshmy, P.B. & Gopan, M. (2008). Coir geotextile-packed conduits for the removal of
biodegradable matter from wastewater. Current Science, 95(5), 655–658.
Rajan, A. & Abraham, T.E. (2007). Coir fiber–process and opportunities: Part 1. Journal of Natural
Fibers, 3(4), 29–41.
Rao, G.V & Balan, K. (Eds), (2000). Coir geotextiles—emerging trends. Alappuzha, Kerala: The Kerala
State Coir Corporation Ltd (publishers).
Schurholz, H. (1991). Use of woven coir geotextiles in Europe. Coir, 35(2), 18–25.
128

Hydrodynamic behavior of Vizhinjam port using Delft3D
E.K. Rajan Shari & N. Sajikumar

Department of Civil Engineering, Government Engineering College Thrissur, Thrissur, India
ABSTRACT: A harbor is a water area partially enclosed by breakwaters and thus pro-
tected from storm-generated waves and currents. It also provides safe and suitable accommo-
dation for vessels seeking refuge, supplies, refueling, repairs, and transfer of cargo. Because
of these protective structures, only small local waves or generated waves are active, but in
a weakened form. The discharge entering into the port basin depends on tidal streams and
weather conditions. In the near-bottom layer of a port, the water movement greatly depends
on the discharge energy in the upper layers, as well as on the changes in the water flow caused
by roughness. These water flows are mainly connected with the displacement speed of the
sediments, and may adversely affect the smooth operations of the port. Hence a thorough
assessment of hydrodynamics is necessary to evaluate the effect of a port development on the
prevailing currents. The proposed port of Vizhinjam, India, was studied in this paper using
the Delft3D-FLOW module for the formulation and modeling of hydrodynamic behavior.
This study mainly analyzed the pattern of water flux entering and leaving the basin area in
different seasons; this information can be used to help in the future planning of the port. The
study results indicated that the breakwater considerably reduced the amount of water in and
out of the port in any situation, and also the wave height within the harbor area.
Keywords: breakwater, bathymetry of port, water flux
1 INTRODUCTION
Kerala in India, lies between the Arabian Sea in the west and the Western Ghats in the east. Ker-
ala’s coast runs 580 km in length, while the width of the state varies between 35–120 km. Large
stretches of this coast are subjected to repeated erosion yearly during the monsoon period.
Additionally, the coastal area is subjected to geological problems and natural processes, such as
coastal erosion, deposition, sedimentation, tsunami and tidal waves. Sediment transport, asso-
ciated with the movement of water, is the main reason for coastal erosion. Breakwaters are usu-
ally built parallel to the shore or at an angle to reduce serious wave action from its destructive
impact on the shoreline. Harbors are created with the help of breakwaters to provide a tranquil
environment, which in turn, blocks the natural movement of water and sediment.
The proposed port at Vizhinjam is located in in the state of Kerala,16 km south of the state
capital Thiruvananthapuram, which falls in proximity to the international east-west shipping
route. Vizhinjam port aims to berth large container ships and hence enhance India’s ability to
handle gateway and transshipment cargo, while establishing a strong supply chain network in
Kerala (Vizhinjam International Sea Port Limited, 2016). The proposed port has a breakwater
that will provide a tranquil environment within the harbor area. Wind and wave-induced flows
will play an important role as the basin is situated along the coast. Analysis of observed silta-
tion rates in various environmental conditions showed that harbor siltation in freshwater con-
ditions is much less than that in salt and brackish water conditions (Nasner, 1992); this may
be due to a high discharge entering the basin which is the combined effect of the tide, waves,
winds and the other physical parameters. Hence it becomes imperative to study the movement
of water within the harbored area so that its effect on sediment transport can also be studied.
129

Thus, the objective of the study is to examine the pattern of water flux entering and leav-
ing the Vizhinjam port in different seasons; this information can be used to help in the future
planning of the port. This study will also be helpful for further studies, such as the sedimen-
tation analysis of harbor basins. For this study, the mathematical modeling tool, Delft3D,
was used to investigate the effect of the breakwater on the rate of flow of water in and out
of Vizhinjam port.
2 METHODOLOGY
Delft3D is a unique, fully integrated computer software suite for a multidisciplinary approach.
Delft3D-FLOW solves the Navier Stokes equations for an incompressible fluid, under the shal-
low water and the Boussinesq assumptions. In the vertical momentum equation, the vertical accel-
erations are neglected, which leads to the hydrostatic pressure equation. In 3D models, the vertical
velocities are computed from the continuity equation. The set of partial differential equations,
in combination with an appropriate set of initial and boundary conditions, is solved on a finite
difference grid (Deltares, 2006). Initial water level above the reference plane is specified, uniform
initial conditions for the velocity component is set to zero, and the water level at the outlet is speci-
fied as boundary conditions. As far as the port of Vizhinjam is concerned, water flux entering and
leaving the port area is a major issue since it is planned to accommodate larger container vessels.
The bathymetry of the Vizhinjam port was conducted using Delft3D software. The master plan
layout (including seabed countor) of Vizhinjam port is as shown in Figure 1. The application
Figure 1. Master plan layout for the proposed

Vizhinjam port. Figure 2. Bathymetry of the proposed Vizhin-
jam port.
Figure 3. The open boundary of the proposed port area.
130

of Delft3D required the systematic input of data, which was generated by using georeferenced
maps. The initial bathymetry was created by using Delft3D Dashboard, Arc GIS and QUICK-
LIN engines in the Delft3D software (Figure 2).
In FLOW GUI, the two physical parameters of wind and wave were provided. The open
boundary condition and astronomical tidal cycles were generated by using the Delft3D
Dashboard (Figure 3).
The morphological file was created for the analysis. To acquire the salient results, three
cross sections were provided. The cross sections were provided at the entry, near the turning
circle, outside the breakwater, and in the approach channel cross section. The simulation was
carried out for 15 months. Initially, it was carried out for 12 months, but the results showed
a cyclic nature. In order to confirm the results, the model was run with 15 months’ worth of
data. However, the simulation time depended on the time step, which is normally restricted
by the Courant stability criteria.
3 RESULT AND DISCUSSION
The flux entering the harbor basin was the first to be analyzed. At the entry cross section,
the discharge was found to be higher from June to July, and lower from February to May.
Figure 4 shows the discharge entering the harbor entry cross section. Q1 and QQ2 represent
quarter 1, quarter 2, respectively, and so on. The maximum discharge in the entry cross sec-
tion was 2.5 × 105 m3/s. The discharge entering the port varied with season. Figure 4 shows
the discharge entering the cross section near the entrance of the harbor. The periodic vari-
ation in dischargewas in accordance with the seasonal variations; the south-west monsoon
being the period in which the major proportion of rainfall occurs. Since no measured infor-
mation is available regarding the amount of discharge in this proposed project, the same was
compared with the available information from the study (Vizhinjam International Sea Port
Limited, 2016). It was found that the results were reasonably matching.
Figure 5 shows the water flux near the turning circle cross section. It can be seen that flow
entering the ship berth area was low compared to other areas. But outside the breakwater
(1.5 km away), the discharge was very high. The maximum and minimum for each case was:
2.2 × 105 m3/s versus 6 × 104 m3/s in the case of the maximum; and −1.2 × 105 m3/s to –3.8 × 105 m3/s
in the case of the minimum. This difference shows the effectiveness of the breakwater system.
Figure 6 shows the discharge across the cross section outside the breakwater. Here also,
the discharge (18 × 106 m3/s) was found to be greater from June to July. Across the cross sec-
tion of the approach channel, the discharge was greater from June to July (6 × 104 m3/s). The
comparison of Figure 6 and Figure 5 shows the role of the breakwater in the amount of
water flux entering the port. The discharge at the cross section inside the turning circle was
reduced to a greater extent.
Figure 5. Discharge in the cross section near

Figure 4. Discharge entering the cross section
the turning circle.
near the entrance of the harbor.
131

Figure 6. Discharge entering the cross section outside the breakwater.
Table 1. Average wave height with season.
Average wave Average wave

Time height (inside) height (outside)
Q1-2016 0.23 0.59

Q2-2016 0.46 0.78
Q3-2016 0.57 1.23
Q4-2016 0.42 0.85
In order to assess the effect of the breakwater on the tranquility of the inside area, the aver-
age wave height for each season at the turning circle cross section, and outside, was measured.
The results are presented in Table 1. It can be seen that the average wave height was reduced by
50% and hence effective in protecting the area. However, the wave height was less than 0.57 m,
which is acceptable for a harbor area
4 CONCLUSION
The hydrodynamic behavior of Vizhinjam port modeled using Delft3D, enabled this study to
evaluate the water flux that was occurring within and around the harbor area. The breakwa-
ter considerably reduced the amount of water in and out of the port in any situation, and also
reduced the wave height. The tranquility of the port area was assessed by estimating the wave
height inside the harbor, and the seasons were found to be of influence. Hence, the current
alignment can provide an environment suited for berthing large ships.
REFERENCES
Deltares. (2006). Delft 3D-FLOW manual. Simulation of multi-dimensional hydrodynamic flows and
transport phenomena, including sediments. Rotterdamseweg, The Netherlands: Author.
Nasner, H. (1992). Siltation in tidal harbours, part I. Die Küste, 127–170.
Vizhinjam International Sea Port Limited. (2016). Vizhinjam international multipurpose sea port project
report.: Author.
132

Traffic and pavement engineering

Operating speed of different classes of vehicles at summit vertical

curves on two-lane non-urban roads
M.J. Neena, M. Harikrishna & M.V.L.R. Anjaneyulu

Department of Civil Engineering, National Institute of Technology Calicut, Calicut, Kerala, India
ABSTRACT: The safety of a road is closely linked to variations in the speed of vehicles
traveling on that road. Horizontal and vertical alignments of a highway are designed based
on an assumed design speed. Later researchers recognized that drivers select a speed influ-
enced by the roadway condition rather than an assumed design speed. This work attempts
to develop operating speed models for crest vertical curves on two-lane non-urban roads in
India. Traffic data, crash data for six years and geometric details of 55 sites formed the data-
base for the study. Scatter plot and correlation analyses were carried to identify the candidate
variables for developing the models to predict operating speed. Multiple linear regression
models were developed. The models developed in the study can be used for the evaluation of
the design of vertical curves and establishing speed limits on non-urban roads.
1 INTRODUCTION
Road crashes inflict heavy economic losses on a country. Safety on roads is necessary to reduce
crashes involving both humans and vehicles; subsequently making roads safer and user-friendly
for traffic. A good geometric design provides an appropriate level of mobility for drivers while
maintaining a high degree of safety. An inconsistency in design will lead to variations in speed.
Operating speed is the most common and simple measure of design consistency. It is charac-
terized as the speed chose by the driver when not discouraged by other users and is normally
represented by the 85th percentile speed, which is denoted as V85 (HSM, 2010).
The vertical grades or curvature of vertical curves of roadways are also related to road
safety. A vertical curve is used to avoid a sudden change of direction when moving from one
grade to another (Lamm et al., 1999). These authors found that a statistically significant
relationship exists between mean speed reductions and mean road traffic crashes; sites with
higher speed reductions showed higher crash rates. Thus, the change in vehicle speed is a
visible indicator of inconsistency in geometric design. Hence, this study attempts to develop
operating speeds on crest vertical curves on non-urban roads.
2 LITERATURE REVIEW
Most research in the field of road safety has focused on establishing relationships between speed,
geometric and operational factors of road segments. This is mainly accomplished by developing
operating speed models that predict the expected speed of a vehicle with specific attributes. Operat-
ing speed is a good indicator of the level of safety on a road segment (Masaeid et al., 1995; Abbas
et al., 2010). Design inconsistencies are quantified by computing the difference between the design
speed and operating speed on a single element or by the difference in operating speeds on two suc-
cessive elements. Generally, the greater the difference between the operating speed and the design
speed of the curve, the greater the design inconsistency and the risk of accidents on the curve.
Jacob et al. (2013) developed operating speed and consistency models for horizontal curves
on non-urban two-lane roads. Hassan (2003) concluded that the grade of the curve and sight
distance are the main concerns in the design of vertical curves. Jeesen et al. (2001) investigated
135

the influence of roadway, traffic and speed characteristics and concluded that posted speed has
the maximum influence on operating speed on rural two-lane crest vertical curves. The impact
of grade, stopping sight distance and speed of vehicle create operational issues for vehicles
traveling on vertical curves and in turn, may increase crashes. Establishing performance-based
safety goals and objectives has become more attainable with the Highway Safety Manual.
According to the Highway Safety Manual (HSM) of the American Association of State High-
way and Transportation Officials (AASHTO) (2010), the factors which affect crashes on verti-
cal curves are driver characteristics, vehicle factors, roadway and environmental factors.
Linear and multiple model forms are commonly employed for predicting vehicle speed
(Jacob et al., 2013; Lamm et al., 1999). The majority of the operating speed models reported in
the literature mostly consider passenger cars. Studies on the influence of geometry on speeds of
other classes of vehicles like heavy vehicles and two-wheelers, which contribute significantly to
the mix of traffic in India, is much less reported. Thus, this paper focuses on the development
of models for predicting the speed of different classes of vehicles on crest vertical curves.
3 METHODOLOGY AND DATA DESCRIPTION
Vertical curves, which are provided to make a smooth and safe transition between two grades,
may or may not be symmetrical. They are parabolic and not circular like horizontal curves.
Vertical curves are of two types: crest vertical curves and sag vertical curves. The length of
the crest vertical should be enough to provide a safe stopping sight distance. Identifying the
proper grade and the safe passing sight distance are the main design criteria in analyzing the
safety aspects of the vertical curve.
The essential part of developing operating speed models for crest vertical curves on non-
urban roadways is data collection. Data required for this study are geometric data and traffic
data along the study sites. Traffic data such as volume count and speed are collected using a
portable infrared based traffic data logger, a user-friendly instrument. For each vertical curve,
traffic data are collected at the Summit Point (SP) and approach tangent or level section, as
shown in Figure 1. Traffic data were collected for 12 hours (6.00 am to 6.00 pm).
3.1 Study sections

The study sites were selected from five districts of Kerala state, India. National Highways
(NH-18, 544, 766, 966), State Highways (SH-1, 22, 24, 28, 34, 48) and some Major District
Roads (MDRs) in each district were chosen.
Figure 1. Traffic data collection location at crest vertical curve.
Table 1. Summary statistics of geometric variables of crest vertical curves.
ATL CW-AT CW-SP EW SD G+ G- CL

ADT (m) (m) (m) (m) (m) (%) (%) (m) K
Minimum 5557 100 6.5 7 0.2 20 2 -6.65 40 12

Maximum 20478 500 10 11.2 2.1 265 6.2 -4.2 568 100
Average 13242 267 7.3 8.1 0.7 114 3.4 -0.9 253 48.8
Std. Dev. 9389 115 1.0 1.2 0.6 60 0.9 3.0 124 21.9
136

The study stretches were identified based on the following criteria. They should be outside
urban areas and be free from the influence of intersections. No physical features or activities, such
as narrow bridges, schools, factories or recreational parks, adjacent to or in the course of the road-
way that may create an abnormal hazard should be present. There should be sufficient shoulder
width for setting up data collection equipment with a tangent length equal to or more than 100 m.
Geometric details of 55 vertical curves were collected using Total Station. The required
geometric particulars are Approach Tangent Length (ATL), Carriageway Width at Approach
Tangent (CW-AT), Carriageway Width at Summit Point (CW-SP), Extra widening (EW),
Sight Distance (SD), Approach Gradient (G+), Departing Gradient (G−), Curve Length
(CL). Change in grade (A) and rate of vertical curvature (K) were collected. Summary statis-
tics of the geometric data and Average Daily Traffic (ADT) are given in Table 1.
The rate of vertical curvature is a measure of the abruptness in a change in grade and influ-
ences visibility on a summit curve. It is defined as the ratio between curve length and change
in grade (K = CL/A).
4 DATA ANALYSIS
Of the 55 data sites, data from 45 sites were used for the model calibration and ten for the
validation of the model.
4.1 Traffic volume

The data collected indicate that the traffic stream mainly consists of cars and two-wheelers.
Around 70% of the traffic volume is contributed by these two vehicle types, followed by Multi
Commercial Vehicle (MCV) that include buses and small trucks. Figure 2 shows the proportion
of different categories of vehicles based on 24-hour traffic volume data at a site on the NH 212.
4.2 Operating speed

Operating speed, which is defined as the 85th percentile speed, was found from the cumula-
tive percent frequency plots for each category of vehicle. Operating speed values of different
vehicle types for the selected site are shown in Figure 3.
Figure 2. Distribution of traffic volume by vehicle type.
Figure 3. Operating speeds of different class of vehicles.
137

Figure 4. Scatter plots for geometric variables with operating speed of all vehicles.
Table 2. Correlations of parameters with operating speed of vehicles.
LMV TW MCV HCV
At At At At
Parameters summit summit summit summit
Traffic volume -0.288 -0.224 -0.422** -0.351*

Approach tangent length 0.54** 0.516** 0.727** 0.580**
Carriageway width AT -0.26 0.376* 0.144 -0.146
Carriageway width SP -0.09 0.477** 0.245 0.269
Extra widening 0.2 0.434** 0.297 0.202
Sight distance 0.02 0.610** 0.321* 0.16
Approaching grade -0.68** -0.394** -0.649** -0.678**
Departing grade -0.27 -0.795** -0.407** -0.177
Curve length -0.34* -0.649** -0.525** -0.480**
Rate of curvature -0.2 -0.211 -0.302* -0.319*
4.2.1 Analysis of variance of speed

The analysis of variance test, a statistical test, was conducted to compare the speed of vehi-
cles of different classes. The test was carried out at level and summit points at 5% significance
level. It was found that there was a significant speed difference between Light Motor Vehi-
cle (LMV) Two Wheeler (TW), LMV – Multi Commercial Vehicle (MCV), LMV – Heavy
Commercial Vehicle (HCV), TW – MCV and MCV – HCV at both level and summit points.
Therefore, all these classes of vehicles were considered for modeling separately. F igure 4
shows the scatter plots prepared for identifying the relationship between explanatory vari-
ables and operating speed of all vehicles. The scatter plots show that the operating speed
increases with an increase in ATL. The operating speed tends to decrease with an increase in
approach grade and the length of the vertical curve at summit points.
138

Operating speed of all vehicles at the summit point seems to increase with the increase in
sight distance. Correlation analysis was also carried out to identifying the candidate variables
for model development shown in Table 2.
5 OPERATING SPEED MODELS
The main objective of this work was to develop operating speed models that take into con-
sideration the geometric features of the road way. Multiple linear regressions are a statistical
methodology describing the relationship between an outcome and a set of explanatory vari-
ables. The form of the multiple linear regression models is given in Equation (1).
Yi = a0 + a1X1 + a2X2 + …… + anXn (1)
where, Yi is the operating speed at road section ‘i’; X1, X2 …… Xn are the explanatory traffic
and geometric variables and a0, a1, a2 …… an are regression coefficients. The regression models
were developed using statistical analysis software. Models were developed for operating speed
at two locations, the approach tangent and summit point for vertical curves. A number of
Table 3. Operating speed models of LMV at summit.
Parameter Estimate Std. Error t value Pr(>|t|)
(Intercept) 6.47E+01 4.351 14.881 < 2e-16

ADT -2.22E-05 1.18E-05 -1.8821 0.86251
ATL 3.18E-02 1.05E-02 3.02 0.00385
G+ -3.212 3.61E-01 -8.887 3.76E-12
CL -3.83E-02 8.38E-03 -4.564 2.94E-05
SD 5.831E-02 2.500 2.338 0.02311
Multiple R2: 0.8614, Adjusted R2: 0.8485.

F statistic: 67.11, p value: < 2.2e16.
Table 4. Operating speed models of LMV at level.
(Intercept) 37.959 6.18971 6.133 8.23e-08

ADT -0.0002 0.00010 -2.266 0.027196
ATL 0.0319 0.00846 3.772 0.000383
CW—AT 2.9861 0.71984 4.148 0.000111

F statistic: 8.21, p value: < 1.02e-06.
Table 5. Operating speed models of HCV at summit.
(Intercept) 6.43E+01 4.9311 13.049 < 2e-16

ADT -1.02E04 9.11E05 -1.117 0.268748
ATL 1.05E-02 9.20E-03 1.139 0.259588
SD 8.63E-03 0.33E-02 2.61 0.518e-01
G+ -4.3710 8.85E-01 -4.933 7.62e-06
CL -2.31E-02 5.82E-03 -3.971 0.000207

Statistic: 19.96, value: 2.265e-11.
139

trials were performed and the one which satisfied the t test, p value, R2 value and adjusted R2
was identified. From these models, the final model was selected based on theroot mean square
error and logic of model coefficient signs. The best fit operating speed models selected for dif-
ferent classes of vehicles at summit and level are presented in Tables 3, 4, 5, 6, 7, 8, 9 and 10.
Table 6. Operating speed models of HCV at level.
(Intercept) 29.479 6.2511 4.716 1.56e-05

ADT -0.0001 0.000102 -1.015 0.315516
ATL 0.0241 0.0085 2.829 0.006395
CW-AT 2.7216 0.724 3.744 0.000418
Multiple R2: 0.4326 Adjusted R2: 0.398.

F statistic: 9.633, p value: 2.964e-05.
Table 7. Operating speed models of TW at summit.
(Intercept) 6.36E+01 4.31 14.767 < 2e-16

ADT -4.63E-05 7.96E-05 -0.581 0.564
ATL 9.76E-03 8.04E-03 1.214 0.23
SD 1.76E-02 1.16E-02 1.514 0.136
G+ -4.18 7.74E-01 -5.404 1.39e-06
CL -1.30E-02 5.08E-03 -2.567 0.013

Table 8. Operating speed models of TW at level.
(Intercept) 4.12E+01 5.06 8.139 3.58e-11

ADT -3.91E-05 8.88E-05 -0.441 0.6612
PTL 1.76E-02 7.40E-03 2.376 0.0208
CW-AT 1.461 5.53E-01 2.64 0.0106
MultipleR2: 0.3474, Adjusted R2 –0.30852.

F statistic: 6.355, p value: 0.0008449.
Table 9. Operating speed models of MCV at summit.
(Intercept) 57.84 6.190 9.343 2.18E-11

ADT -0.0005 0.0002 -2.136 0.0392
ATL 0.024 0.010 2.219 0.0325
G+ -2.879 1.209 -2.38 0.0224
CL -0.0022 0.008 -0.28 0.7807
SD 0.0038 0.016 0.242 0.8098

140

Table 10. Operating speed models of MCV at level.
(Intercept) 35.65 7.5625 4.713 2.94e-05

ADT -0.00038 0.0002 -1.821 0.07608
ATL 0.0242 0.0091 2.67 0.01091
CW-AT 2.866 0.7237 3.961 0.00029

Figure 5. Variation of operating speed with curve length and grade.
Table 11. PRMSE values.
PRMSE values PRMSE values
LMV HCV
Calibration Validation Calibration Validation
Tangent 6.97 8.48 9.27 8.54

Summit curve 8.49 10.55 8.48 8.39
PRMSE values PRMSE values

TW Bus
Calibration Validation Calibration Validation
Tangent 8.7 12.33 12.68 10.46

Summit curve 7.96 10.98 10.12 10.08
The coefficients of determination (R2) varied from 0.43 to 0.86 and adjusted R2 (Ra2) from
0.31 to 0.84. All the selected models at the summit curve have F values higher than the
standard table values. Also, the coefficients of the predictor variables are significantly dif-
ferent from zero, as per the t test values at the 95% confidence level. The positive sign of the
141

approach tangent speed (V85 AT) indicates that, if the vehicle arrives with a high speed at the
level, then there will be an increase in the speed at the summit point. Sight distance also has
a positive association with the operating speed except for two-wheelers.
Figure 5 shows the relation of curve length and grade with operating speed based on devel-
oped models, with ADT = 8000, ATL = 100 m, G+ = 3% and Sight Distance = 100 m. It can
be seen that an increase of curve length from 50 to 350 m, reduced the operating speed of all
vehicle classes by about 3 to 11 km/h except for buses. In case of grades, when ADT = 8000,
ATL = 100 M, SD = 100 m and curve length = 100 m, it can be noticed that one unit change
of grade of curve has a significant effect on the operating speed for all vehicle classes.
Data from ten sites were used for calculating the percentage root mean square error
(PRMSE) and validation of models. PRMSE values are shown in Table 11. The PRMSE
values indicate that the models are reasonably good.
This study summarizes the analysis and development of operating speed models for vertical
curves. The analysis was done to determine the variables that significantly influence speeds
of vehicles at vertical curves. Correlation and scatter plot analyses showed that curve length,
approach grade, ATL and sight distance are the variables that affect the operating speed at
limiting and summit points. Grade and curve length are the most important variables for
predicting the operating speed of all categories of vehicle at the summit curve. The operat-
ing speed models will be useful for geometric consistency evaluation of crest vertical curves.
ACKNOWLEDGMENTS
The authors sincerely thank the Center for Transportation Research, Department of Civil
Engineering, National Institute of Technology Calicut, a center of excellence setup under
the Frontier areas of Science and Technology (FAST) Scheme of the Ministry of Human
Resource Development (MHRD), Govt. of India for the support received.
REFERENCES
Abbas, S.K.S., Adnan, M.K. & Endut, I.R. (2010). Exploration of 85th percentile operating speed
model on crest vertical curve two-lane rural highways. Proceedings of Malaysian Universities Trans-
portation Research Forum and Conferences (MUTRFC). University Tenaga Nasional.
Hassan, Y. (2003). Improved design of vertical curves with sight distance profiles. In Transportation
Research Record (TRB) (1851, 13–24). Washington DC: National Research Council.
Highway safety manual, (2010), Washington DC American Association of State Highway and Trans-
portation Officials (AASHTO).
Jacob, A., Dhanya, R. & Anjaneyulu, M.V.L.R. (2013). Geometric design consistency of multiple hori-
zontal curves on two-lane rural highways. Procedia-Social and Behavioral Sciences, 104, 1068–1077.
Jessen, D.R.K.S., Schurr, P.T., McCoy, G. & Pesti, R.R. (2001). Operating speed prediction on crest
vertical curves of rural two-lane highways in Nebraska. In Transportation Research Record (TRB)
(1751, 67–75). Washington DC: National Research Council.
Lamm, R., Psarianos, B. & Mailaender, T. (1999). Highway design and traffic safety engineering hand-
book. Washington DC: Transportation Research Record, Transportation Research Board.
Masaeid, A.L.H.R., Hamed, M., Aboul, E.M. & Ghannam, A.G. (1995). Consistency of horizontal
alignment for different vehicle classes. In Transportation Research Record (1500, 178–183). Washing-
ton DC: Transportation Research Board.
142

Comparison of level of service measures for two-lane rural

highways in mixed traffic
M. Sangeetha, M.V.L.R. Anjaneyulu & M. Harikrishna

Department of Civil Engineering, NIT Calicut, Kozhikde, Kerala, India
ABSTRACT: Two-Lane Highways carry maximum inter and intra-state traffic in any
country. The Level of Service of the facility can either be based on speed or platoon. Out of
the seven LOS measures studied, two platoon-related measures were found to be more suit-
able for LOS analysis of two-lane highways, which are Number of Followers and Follower
Density. The NF and FD are modeled using flow in the same direction and flow in the oppos-
ing direction. It was found that the linear model fits well for the data.
Keywords: Level of service, Platoon, Average travel speed, Percent free flow speed, Number
of followers, Follower density
1 INTRODUCTION
Two-lane highways carry a major portion of passenger and freight transport in any country.
In India, more than 50% of National Highways are of two-lane category (MORTH, 2012).
Functions of National Highways range from carrying high speed inter-state and intra-state
traffic to providing access to most remote parts of the country. Hence, the quality of traf-
fic operation of two-lane highways is a matter of concern while considering the economic
development of the nation.
Two-Lane highways are high-speed single carriageway roads with two-lane, one each for the
use of traffic in each direction. Unlike any other facilities, the overtaking has to be carried out
through the lane meant for traffic in opposite direction. The overtaking opportunity is restricted
by the availability of the gaps in the opposing lane and sight distance. If there is no suitable over-
taking opportunity available in the opposing lane, the fast moving vehicle has to wait behind the
slow moving vehicle, till it gets a suitable opportunity to overtake. With the increase in traffic flow,
more vehicles join the queue. This moving queue, which is otherwise termed as ‘platoon’, causes
reduction of the quality of traffic operation of two-lane significantly. Hence the traffic operation
on a two-lane highway needs to be assessed using suitable Level of Service (LOS) measures.
2 BACKGROUND LITERATURE
Highway Capacity Manual, 2010 (HCM, 2010) published by Transportation Research Board,
USA has suggested that LOS measure should be easy to measure and easy to understand by
users. The quality of operation may be predicted using one or more measures depending on
the requirement. The measures are:
2.1 Speed-related measures

2.1.1 Average Travel Speed (ATS)
It is the average speed of vehicles on a highway segment including the stopped delay. The
value varies from 55 mi/hr (90 km/hr) for LOS A to 40 mi/hr (65 km/hr) for LOS E (HCM,
143

2010). ATS is easy to measure and can relates well with user’s perception (Penmetsa et al.
2015, Hashim & Wahed, 2011, Al-Kaisy and Durbin, 2011 & Al-Kaisy and Karjala, 2008).
But it has a poor relationship with flow rate, with very less R2 value. But it cannot be com-
pared among sites, as it lacks a common reference point.
Average speed of passenger cars (ATSpc) has also been tried as an LOS measure as speed
of passenger cars is more sensitive to traffic level on two-lane highways than that of heavy
vehicles (Penmetsa et al. 2015), Hashim & Wahed, 2011, Al-Kaisy and Durbin, 2011 and
Al-Kaisy and Karjala, 2008). Brilion & Weiser 2006) established a non-linear relationship
between ATSpc and traffic volume while analyzing the LOS in Germany.
2.1.2 Percent Free Flow Speed (PFFS)

Percent Free Flow Speed (PFFS) is another LOS measure (Penmetsa et al. 2015, Hashim &
Wahed, 2011, Al-Kaisy and Durbin, 2011 and Al-Kaisy and Karjala, 2008). PFFS is the abil-
ity of the vehicle to travel at or near the posted speed limit.
PFFS = ATSd/FFS (1)
where, ATSd is the Average Travel Speed in the direction considered and FFS is the free—
flow speed. This can be compared with other sites. If the value is close to 1 it indicates a good
performance condition.
2.2 Platoon-related measures

2.2.1 Percent Time Spent Following (PTSF)
This measure was first introduced by HCM 2000. It is the average percentage of travel time that
vehicle must travel in platoons behind slower vehicles on a two-lane highway, due to the inability
to pass. This spatial measure is difficult to collect from field. PTSF also represents the approxi-
mate percentage of vehicles travelling in platoons. The platoons are identified based on critical
time headway. HCM suggests this critical headway as 3 seconds. Therefore, percentage of vehicles
moving with a headway less than or equal to 3 seconds was considered for formulating PTSF.
2.2.2 Percent Follower (PF)

A follower is any vehicle in the traffic stream, following another vehicle, in the same lane by
relatively short headways (Al-Kaisy and Karjala, 2008). Percent Follower (PF) represents the per-
centage of vehicles with short headways in the traffic stream. It is easy to measure using surrogate
headway of 3 seconds. The limitation of this measure is that it does not accurately reflect the traf-
fic level, which is an important LOS concept. Sometimes the low traffic level can have high PF if
speed variation is high and passing opportunities are limited (Hashim & Wahed, 2011). This will
give a misleading interpretation in highway up gradation and improvement program.
2.2.3 Follower Density (FD)

Follower Density (FD) is the number of followers in a directional traffic stream over unit
length, typically 1 km or 1 mile.
FD = PF × Density (2)
Density = Flow rate/ATS (3)
where, FD is the Follower Density and PF is the Percent Followers.

The advantage of this measure is that it considers the traffic level also into account.
This can capture freedom to manoeuvre and degree of congestion. This also has the best
correlation with traffic variables (Al-Kaisy and Karjala, 2008). This can be compared
with the ‘density’ in freeways and multilane highways.
2.2.4 N umber of Followers (NF) and Number of Followers as a Percentage of Capacity

(NFPC)
It is the number of vehicles moving in the same lane of two-lane highways with gap less than
2.6 seconds As NF cannot be compared with other sites, NFPC was introduced (Penmetsa
144

et al. 2015). It is the NF expressed as a percentage of capacity. It was found to be the best
measure for LOS analysis in mixed traffic conditions. It has a very good correlation with flow
rate. It reflects the congestion level of the road stretch.
2.3 Platoon identification

Platoons can be identified based on critical headway or critical speed and mean speed differ-
ence between leader and follower vehicle.
3 AIMS AND OBJECTIVES
The aim of the research is to identify a suitable LOS measure, which best reflects the traffic
operations on two-lane highways.
The objectives of the work include identifying the factors influencing LOS of two-lane
highways, estimating and comparing various performance measures for the facility, check-
ing the suitability of the various performance measures and to suggest the best performance
measure which best reflect the actual field condition.
4 METHODOLOGY AND DATA COLLECTION
4.1 Methodology
The speed related measures estimated are Space Mean Speed, Standard deviation of Space Mean
Speed and Coefficient of Variation of Space Mean Speed. The platoon-related measures used
include Number of Followers, Percent Followers, Follower Density and Average Platoon Size.
Data is collected from 11 straight and level sections of various districts of Kerala. The data col-
lected include traffic volume, speed and time headway of individual vehicles. The traffic data is
collected using an infrared based automatic traffic logger and video camera. The various speed-
related measures such as Space Mean Speed, Standard deviation of Space Mean Speed and Coef-
ficient of Variation of Space Mean Speed and various platoon-related measures like Number of
Followers, Percent Followers and Follower Density are estimated and plotted against 5 minute
traffic flow rate. All the performance measures were estimated for all the 11 sections separately.
In order to study the variation of the LOS measures due to traffic factors, the sites are
selected from sections of uniform geometry. The uniform, level sections of roads of varying
length were selected from National Highways, State Highways and Major District Road in
Kozhikode, Malappuram, Thrissur and Palakkad.
The traffic data required for estimating various speed and platoon related measures was
collected at two-levels. One set of data was collected using video recording. This data was
used to identify the vehicles in platoons. Other set of data was collected using automatic traf-
fic logger. The data was collected on a clear weather day for a period of 10 to 12 hours from
6 am to 6 pm. The results of the first set of data were used to analyse the second set of data
and establish various relationships between them. Figure 1 shows the site selected for vide-
ographic survey (left) and the automatic traffic logger (right).
4.2 Data collection

The video graphic data was collected in order to arrive at threshold headway, which can be
used to define the vehicles in platoon. The video camera was kept at elevated point away from
the road. Platoons were visually identified and headway of platooned vehicles was extracted
from the video data.
The automatic traffic logger is used to collect the data at 11 sections of road. The infra-
red based equipment captures the vehicle class, spot speed, direction of travel, lateral posi-
tion across the width of the road and time headway. The instrument set up is shown in
Figure 1.
145

Figure 1. Site for videographic survey (left) Traffic logger (right).
5 DATA ANALYSIS
5.1 Estimation of speed-related measures

For a uniform section of road, ATS can less accurately expressed by using space mean speed.
It is very easy to estimate too. Therefore, SMS was the first LOS measure estimated. The
speed-related measures estimated include space mean speed, standard deviation of Space
Mean Speed and coefficient of variation of space mean speed.
5.1.1 Space Mean Speed (SMS)

Space Mean Speed was calculated for 5 minute interval. The estimated values are plotted
against 5 minute flow rate, considering both directions.
5.1.2 Standard deviation of speed

The standard deviation of the speed with flow rate is analysed by using this relation. The
relationship is observed to be a weak one. The measure cannot be used as LOS measure.
5.1.3 Coefficient of variation of speed

Coefficient of variation is the ratio of standard deviation to mean. It clearly represents the
flow condition at site. The lower coefficient of variation represents a better flow condition.
The Figure 2 shows the relationship between speed-related measures and traffic flow rate. The
relationship obtained was weak, which shows that they cannot be used as a good LOS measure.
5.2 Estimation of threshold headway

The data collected using video recorder was used to estimate the threshold headway. The 95th
percentile of time headways is taken as the threshold headway. The 95th percentile value was
obtained as 3.25 seconds. Figure 3 shows the estimation of threshold headway.
5.3 Estimation of platooning-related measures

The various platooning related measures analysed include NF, PF, FD and APS.
5.3.1 Number of Followers (NF)

It is the number of vehicles that follow behind the slow moving vehicles in 5 minute interval. The
‘following’ vehicles are identified from the whole data base using 3.25 seconds as threshold head-
way. The result shows that the NF has a good relation with that of flow rate. The combined data
of the 11 sections were found to fall in 3 categories. This may be due to the variation in geometry.
5.3.2 Percent Followers (PF)

It is the percentage of vehicles which travel with headway less than 3.25 seconds. The relation-
ship does not hold well when compared to that of NF.
5.3.3 Follower Density (FD)

Follower Density (FD) is the number of vehicles following the per unit length of road stretch.
FD is a good measure as NF. Another advantage of using FD as an LOS measure is that
146

Figure 2. Speed-related measure and flow rate.
Figure 3. Estimation of threshold headway.
Figure 4. Platoon related measures with flow rate.
Table 1. Modelling result of NF.
Location R square Standard error Observations A B C
1 0.824 2.38 126 -6.98 0.04 0.004

-8.46 19.65 2.25
2 0.724 2.927 126 -11.06 0.04 0.009
-6.616 17.8 3.82
3 0.879 5.124 144 -10.42 0.06 -0.01
-5.77 26.4 -3.11
4 0.922 3.701 120 -20.07 0.07 0.006
-10.45 36.3 2.81
5 0.889 4.43 139 -16.41 0.05 0.003
-10.04 32.9 2.55
it represents the traffic over a stretch. Similar to NF, FD also falls into three categories.
Aamong all the measures, NF and FD have best relation with flow rate, which means, NF
and FD can be used as the LOS measures for two-lane highways. Figure 4 show the relation-
ship between platoons related measures and flow rate.
Out of seven performance measures, which include four speed-related measures and three
platoon-related measures, FD and NF are found to have a good correlation with traffic flow.
The combined data from 11 locations fit trend lines which indicate the effect of factors other
than traffic on these measures.
147

The NF and FD are modeled using flow in the same direction and flow in the opposing
direction. It was found that linear model fits the data well. Model form is as shown below,
where A,B and C are constants.
LOS measure = A* x1 + B* x2 + C (4)
where, x1 = Flow in the same direction

x2 = Flow in the opposing direction
Table 1 shows the modeling results at five sites.
The study is based on the traffic data from 11 straight and level sections of two-lane highway.
Four speed-related measures and three platoon-related measures were analysed by plotting
the same with respect to 5 minute flow rate. Out of the 6 measures, Number of Followers
(NF) and Follower Density (FD) are found to be more suitable measures for LOS analysis of
two-lane highways. Linear model found to fit well for the data.
The major cause for the reduction in the quality of operation of two-lane highways is the
forceful following of fast moving vehicles behind the slow moving vehicles due to the inabil-
ity to overtake. The ‘following’ feature is an important aspect on two-lane highway, which
decides the LOS. NF and FD are two measures which incorporate this particular aspect of
the traffic operation on two-lane highway. While NF considers the number of followers at a
point on the section of road, FD considers the number of followers per unit length of road.
The measures are found to be affected by the geometry too.
The main limitation in the study is the assumption made on ‘following’ criteria. It is
assumed that all vehicles have same ‘following’ pattern, and same threshold headway of
3.25 seconds. But, it is likely to vary depending on factors like vehicle class, driver and local-
ity. NF is a number. Comparison among the sites may not be possible as it depends on many
other factors as mentioned above.
ACKNOWLEDGEMENT
We are thankful to the Centre for Transportation Research, Department of Civil Engineering
for providing the necessary infrastructure for conducting the research work.
REFERENCES
Al-Kaisy, A. and Durbin, C. 2011. Platooning on two-lane two-way highways: an empirical investiga-
tion. Procedia—Social and Behavioral Sciences, 1(16), pp. 329–339.
Al-Kaisy, A. and Karjala, S. 2008. Indicators of performance on two—lane rural highways. Transporta-
tion Research Record, 2071, pp. 87–97.
Brilon, W. and Weiser, F. 2006. Two-lane rural highways: the German experience. Transportation
Research Record: Journal of the Transportation Research Board, No. 1988, Transportation Research
Board of the National Academies, Washington, D.C., 2006, pp. 38–47.
Hashim, I. and Abdel-Wahed, T.A. 2011. Evaluation of performance measures for rural two-lane roads
in Egypt. Alexandria Engineering Journal, No. 50, pp. 245–255.
MoRT&H (Ministry of Road Transport and Highways). 2012. Basic Road Statistics of India 2008–09,
2009–10 & 2010–11 Government of India, New Delhi, India.
Oregon Department of Transportation 2010. Modelling Performance Indicators on Two-Lane Rural
Highways: The Oregon Experience.
Penmetsa, P., Ghosh, I., and Chandra, S. 2015. Evaluation of performance measures for two-lane inter-
city highways under mixed traffic conditions, Journal of Transportation Engineering, 141(10), pp. 1–7.
Transportation Research Board, Highway capacity manual. 2010, USA. 2. Chapter 15.
148

Effect of rural highway geometry on driver workload:

A step towards safety
Anitha Jacob, K.J. Jinesh, Jisha Akkara & Jose P. Therattil

Jyothi Engineering College, Cheruthuruthy, Thrissur, Kerala, India
ABSTRACT: The research on traffic safety on highways underlines the need for maintaining
consistency in the geometric design of highways. This paper focuses on driver workload and
how the geometry affects driver’s physiological characteristics. The work presented includes
development of a device namely, the Road Driver Data Acquisition System (RDDAS). It is
comprised of various sensors for capturing heart rate and galvanic skin resistance, and video
cameras for capturing eye blink rate. Drivers equipped with the RDDAS were allowed to
drive a vehicle with Global Positioning System through study stretches of known geometry.
The effect of geometry on driver workload was explored using a scatter plot study and cor-
relation analysis. The results indicate that heart rate and rate of eye blinking are very good
indicators of driver workload. The study could be extended to develop mathematical models
that can quantify the relationship between highway geometry and driver workload.
Keywords: driver workload, highway geometry, heart rate, galvanic skin resistance, rate of
eye blinking
1 INTRODUCTION
The influence of geometry has an upper hand in controlling vehicular movement in rural
(non-urban) highways where speed matters much more than traffic volume. Highways
through rural areas generally support intercity trips. Drivers read the road ahead of them
and adopt a speed that seems comfortable to them. Any unexpected road feature in the high-
way may surprise the driver and lead to erroneous driving maneuvers which in turn may end
up in crashes. As highways are meant for high speed travel, the impact of any collision that
occurs will be of a grievous nature. Hence, highways need to be designed in such a way that
the geometry itself guides a driver to adapt a maneuver fitting the environs.
To improve traffic safety, designers and planners use many tools and techniques. One
technique used to improve safety on roadways is geometric design consistency. Design con-
sistency refers to the highway geometry’s conformance with driver expectations. Generally,
drivers make fewer errors at geometric features that conform to their expectations.
There are several measures for evaluating the consistency of geometry. These measures are
classified as speed-based measures, alignment indices, vehicle stability-based measures and
driver workload measures (Fitzpatrick et al., 2000). Among the measures, the driver work-
load measure is the sole method which directly considers the effect of geometry on drivers. As
drivers are the major road users, it is always logical and appropriate to evaluate a road design
from the view point of its major beneficiaries. Workload will be increased as and when the
mismatch between what a driver observes in the field and what he expects increases. When this
inconsistency increases beyond a limit, the driver may adopt an erroneous driving maneuver
and this may result in a crash (Messer et al., 1981; Kannellaidis, 1996). Maintaining design
consistency minimizes driver workload and thereby reduces the chances of a crash.
This study is part of the research being carried out to derive guidelines for geometric
design based on driver workload. The prime objective of this paper is to focus on the effect
149

of the geometry of two-lane rural highways on driver workload. This is achieved through
two steps—data collection and data exploration. The former includes the development of
a system for collecting and storing driver workload data and the conduct of an exhaus-
tive geometric survey. Subsequently, the correlation between geometric variables and driver
physiological condition is explored and quantification completed.
2 BACKGROUND
At present in India, a highway is designed based on a design speed concept. Highway design-
ers design different elements of highways based on a standard value of design speed and on
the premise that drivers will not exceed the speed limit set below the design speed for traffic
safety. However, speed studies have always pointed to the fact that drivers adopt their speeds
according to the highway ahead and may exceed both the speed limit and the design speed
(Leisch & Leisch, 1977). This mismatch has been attributed mainly to the fact that design
speed theory does not consider the effect of highway geometry on driver’s performance and
the variations within and between design elements.
Driver performance can be evaluated based on the workload involved while performing
the driving task. The driver is almost continuously gaining visual and kinaesthetic informa-
tion, processing it, making decisions and performing accordingly. This includes tracking the
lane, speeding up and down as demanded by the geometry, steering the vehicle and many
other tasks (Brookhuis & de Waard, 2000). A driver can only adapt to the road system as he
gathers most of the information from the road itself. The various characteristics that contrib-
ute to workload are vertical and horizontal alignment, consistency of road elements, roadside
obstructions, presence of other vehicles and environment (Weller et al., 2005). Thus, a road
system should be designed as “user friendly” in order to reduce driver errors. Hence, rating of
a highways based on driver workload will be more commensurate with reality.
Perception of highway alignment, especially complex or hazardous ones can induce stress
on drivers. Thus, driver workload is highly correlated with accidents. If workload drops too
low or rises too high, the collision rate can increase. If workload is too low, drivers may
become inattentive or tired, and their responses to unexpected situations may then be inap-
propriate or slow. At the other end of the scale, drivers may become confused by very high
workload. In these situations, drivers may overlook or misinterpret an unexpected occur-
rence and either not respond until too late or respond inappropriately. This is closely related
to the arousal law that takes the form of an inverted U-shaped function between arousal level
and performance (Yerkes & Dodson, 1908).
There are five broad categories of workload measurement techniques (Green et al., 1993)
including primary task measurement, secondary task measurement, physiological measure-
ments, subjective techniques and input control. In this study, a physiological measurement
technique is made use of to record driver workload.
Several non-invasive physiological measurements are thought to measure aspects of opera-
tor state that correlate with the ability to perform tasks. Measures explored include heart rate
and heart rate variability (Mulder, 1980) and eye-blink rate (Stern, Walrath & Goldstein,
1984). Heart rate tends to increase with an increase in workload. The rate of eye-blink can
be used as a measure of visual demand. Eye-blink rate decreases with an increase in visual
demand. Other measures are galvanic skin conductance, pupil diameter, blood pressure, res-
piration rate, hormone levels and electromyogram. Helander (1978) studied the effect of nar-
row roads and narrow shoulders on heart rate and electrodermal response of drivers and
found that perception of a complex road way induces stress in drivers. The method of assess-
ing workload through physiological measures has been found promising, though it requires
devices that can capture the physiological measures.
The literature review showed that many researchers are trying to arrive at a methodol-
ogy for quantifying the workload of drivers by means of real driving task measurements or
through simulator studies. Studies show that an abrupt increase in driver workload increases
the probability of crashing and such increases could be coined with many roadway features
150

(Krammes et al., 1995; Cafiso et al., 2005). Messer (1980) evaluated workload as a function
of familiarity of driver, feature expectancy, sight distance and workload of previous feature.
It was concluded that the workload is higher during encounters with complex geometric fea-
tures and can be dramatically higher when drivers are surprised by encounters with combina-
tions or sequences of severe geometric features.
In 1989, Hulse et al. proposed a general model (Equation 1) to quantify driver workload.
This model takes into account the sight distance, curvature, obstructions closer to the lane
and road width.
Q = 0.4 A + 0.3B + 0.2C + 0.1D (1)
where, A = sight distance factor; B = curvature factor; C = lane restriction factor; D = road
width factor.
Green et al. (1993) further examined the Hulse model and related workload to geometry
and found that the standard deviation of lateral positioning of the vehicle is negatively
correlated to workload. The study showed that sight distance deficiency increases driver
workload.
Krammes et al. (1995) defined workload as the portion of the total driving time that driv-
ers need to look at the roadway and measured it using the vision occlusion method. They
expressed workload in terms of the degree of curvature as given in Equation 2.
WL = 0.193 + 0.016 D (2)
where, WL = workload; D = degree of the curve.

Using the visual occlusion technique, Wooldridge et al. (2000) measured a driver’s visual
demand on horizontal curves as a percentage of the time a driver observes the roadway. Ana-
lytical models that relate visual demand to the inverse of the horizontal curve radius were
developed using data collected from a test track, local road and simulator.
Cafiso et al. (2005) used a driver instrumented vehicle acquisition system for collecting
actual driving data for the road and driver. The study confirmed that an isolated curve fol-
lowing a long tangent provoked a driver more than a coordinated sequence of curves. Simi-
larly, a lack of transition curves is a cause of inconsistency.
Suh et al. (2006) conducted field experiments with an instrumented vehicle for tracking
eye movement and found that eye movement is significantly correlated to geometry and
illumination.
A review of literature on driver workload and geometric consistency has ratified that
driver data analysis and evaluation will give original information about the workload of the
driver and its correlation with road features.
The improvement of road safety through geometric design consistency is only an upcom-
ing technique in India. Although the majority of severe road crashes take place on non-urban
highways, very little research has been carried out regarding predicting crashes on rural roads
and still less is oriented to promoting road safety through consistency evaluation of roadway
geometry. Among the various methods available for evaluating consistency, operating speed-
based evaluation is the one most frequently utilized in many research institutions. Only a
small amount of research has been done on workload-based evaluation of consistency of
geometry. Hence, there is a lot of scope for research in this area.
3 METHODOLOGY
A review of the literature showed that driver workload is very much related to the geometry of
the highway section. How this geometry influences driver expectancy is the subject matter of
this paper. The methodology includes three parts related to data collection:- Development
of the Road Driver Data Acquisition System (RDDAS), collection of geometric data and col-
lection of driver workload data.
151

3.1 Development of RDDAS—hardware
The RDDAS is an indigenous system developed and fabricated to measure and records a
driver’s physiological data like heart rate, galvanic skin resistance (GSR) and eye-blink. The
various components include a GPS, two video cameras to record the road and driver char-
acteristics, and physiological data measuring sensors. Figure 1 shows the components and
layout of the RDDAS fitted on a driver. The heart rate sensor is attached to ear. The gal-
vanic skin sensor is attached on palm of the hand. The eye-blink rate is retrieved from the
recorded video of the driver. A vehicle equipped with RDDAS is run through the selected
study stretches by different drivers to collect their workload information.
Figure 1. Road driver data acquisition system.
Figure 2. Algorithm for hardware of RDDAS.
152

3.2 Development of RDDAS—software
A monitoring tool is designed in Visual Studio. Net 2012 to monitor and store various sensor
output and GPS data. The output is taken in Microsoft Excel format. Two different serial
ports are used for acquiring data, one is for GPS and the other for Arduino. A code was writ-
ten in Visual Studio C# to capture the video from two web cams and store the same in a drive.
Figure 2 shows the algorithm for hardware development.
3.3 Geometric data collection

The study sections are selected from two-lane rural highway stretches of Thrissur district.
A geometric survey of 28 horizontal curve sites was carried out by conducting a total station
survey. Geometric data include radius of curve, curve length, deflection angle, degree of
curvature, superelevation, width of roadway and preceding tangent length. The summary
statistics of the data are given in Table 1.
3.4 Driver workload data collection

Each driver was asked to drive a 32 km stretch wearing the RDDAS sensors. The heart rate
sensor was clipped to the ear and a GSR sensor was attached to the fingers of the driver with-
out interfering with the driving. Eye-blink was detected by capturing the driver’s face using a
USB camera fixed on the front glass using a suction cup holder. Another camera was facing
the front side to record the road conditions. All were connected to a laptop along with a GPS
module. Figure 3 shows the driving set up for car and bus.
After the setup is done on the vehicle and driver, the driver is given a rest period of about ten
minutes in order to record his/her normal workload. Subsequently, the driver will complete a
ten kilometer stretch before the start of the study stretch to familiarize the driver with the vehi-
cle and the RDDAS. During the drive along the study stretch, workload data is collected con-
tinuously along with GPS coordinates. The program developed in the study makes it possible to
Table 1. Summary statistics of geometric and driver workload data.
R CL DA W SE PTL 85 AVG 50 85 50 AVG 85 50 AVG

(m) (m) (deg) (m) (%) (m) HR HR HR GSR GSR GSR EB EB EB
Mean 161.12 46.26 18.04 6.63 2.56 144.64 95.54 83.25 83.34 521.89 417.20 394.32 0.59 0.25 0.31
Med. 152.15 38.64 15.50 6.81 2.77 115.00 96.00 82.89 82.75 523.40 415.50 393.06 0.54 0.25 0.31
Mode 157.92 20.56 11.00 6.43 0.53 90.00 96.00 82.43 81.50 526.00 415.00 #N/A 0.50 0.25 #N/A
Std. Dev 71.07 20.92 11.19 0.73 1.85 74.76 2.25 1.84 2.99 7.47 9.42 7.03 0.19 0.11 0.09
Min. 33.38 20.56 7.00 4.94 0.03 60.00 91.10 78.92 78.00 501.20 400.00 374.68 0.32 0.00 0.13
Max. 303.47 94.63 58.00 7.83 7.77 370.00 99.50 86.87 91.00 532.25 437.00 407.56 1.00 0.50 0.53
Figure 3. Driver wearing RDDAS device.
153

tag the information regarding start and end of each isolated horizontal curve along the route.
This helps to quantify the workload data corresponding to each curve. The output data include,
an Excel sheet showing GPS and workload data and the video footage of the driver and road
are saved onto the laptop with the date and name of the driver. The drives were completed for
28 horizontal curves with 32 car drivers. Table 1 shows the summary statistics of geometric and
driver workload data. Three parameters were taken as representative of driver data including
heart rate, galvanic skin resistance and rate of eye blinking. Workload data were assigned to
each curve by taking their 85th percentile value, 50th percentile value and average value.
All the highway sections were of a two-way two-lane category with the width of road vary-
ing from 4.94 m to 7.83 m. The surface condition was good during the driving exercise. All
driving experiments were done during dry weather, off-peak periods to minimize the influ-
ence of climate and traffic volume on workload. The total time for the round trip was around
one hour and forty five minutes. Before the return trip, the driver was given a sufficient rest
period so that fatigue was not affecting his driving behavior.
3.5 Data analysis

The sample size of the study was 28. The data of the 28 curves include geometric data and
workload data. Figure 4 to Figure 9 show scatter plots of the variation of workload data
with the geometry of the curve. Of the various geometric variables considered, the radius and
deflection angle were found to have a significant influence on driver heart rate, skin resistance
and eye blinking.
A critical study of various scatter plots shows that geometry has a significant effect on work-
load. The sharper the curve, the higher the driver workload which is indicated by higher heart
Figure 5. Effect of radius on GSR.

Figure 4. Effect of radius on heart rate.
Figure 6. Effect of superelevation on GSR. Figure 7. Effect of preceding tangent length on

GSR.
154

Figure 9. Effect of deflection angle on rate of
Figure 8. Effect of radius on rate of eye
eye blinking.
blinking.
Table 2. Correlation of geometry and workload.
R CL DA W SE PTL
85 HR -0.39* -0.15 0.19 -0.10 0.14 0.05

50 HR -0.45* -0.25 0.09 -0.14 0.10 -0.11
AVG HR -0.40* -0.05 0.27 0.27 0.08 -0.11
85 GSR -0.17 -0.15 -0.02 0.11 0.43* 0.21
50 GSR 0.10 0.30* -0.23 -0.06 -0.09 0.37*
AVG GSR -0.27 -0.04 -0.36* 0.17 0.14 -0.13
85 EB -0.09 -0.29 -0.36* 0.13 -0.18 -0.19
50 EB 0.06 -0.27 -0.31* 0.20 -0.21 0.03
AVG EB 0.01 -0.20 -0.28 0.09 -0.16 -0.14
*Significant correlation.
rate, lower galvanic skin resistance and less eye blinking. At sharper curves, the driver needs to be
more attentive which makes the heart work harder. With greater workload, sweating increases,
thereby reducing skin resistance. Also, the driver needs to keep his eyes open for more time to
negotiate the curve. As superelevation and preceding tangent length to the curve increases, skin
resistance also increases, which is indicative of a reduction in workload. The greater the deflec-
tion of the curve, the lesser the rate of blinking and hence, the greater the driver workload.
A correlation study was performed to understand and quantify the linear correlation between
geometry and workload. Table 2 gives the result of the study. The radius was found to have
negative correlation with the heart rate. The curve length had 30% correlation with the 50th
percentile GSR. The deflection angle had a negative correlation with the average skin resist-
ance and rate of eye blinking. Superelevation and preceding tangent lengths are other variables
which were found to have a correlation with galvanic skin resistance. It was noted that the width
of the pavement did not have any linear correlation with any of the workload measures.
4 CONCLUSIONS
The work done based on 32 drivers’ data on 28 horizontal curves of two-lane rural highways
of Kerala has revealed that geometric design has a significant influence on driver workload.
Radius and deflection angle are two design elements of geometry that were found to have
a higher influence on geometry. Workload was found to increase at curves with a smaller
radius and higher deflection angle. The rate of eye blinking and heart rate are two promising
workload-based measures that can be made use of in estimating the consistency of geometric
design of highways. The work can be further extended to model the relationship between
155

geometry and consistency with a future scope of arriving at threshold values for consistency
evaluation.
ACKNOWLEDGMENTS
The authors acknowledge the Engineering and Technology Program division of Kerala State
Council for Science, Technology and Environment, Trivandrum for their financial support
and assistance. The technical support given by Dr. T. Elangovan., Exe. Director, Road safety,
Kerala Road Safety Authority is greatly acknowledged. Special thanks to the project staff,
Mr. M.R. Midhun and Ms. P. Maheswari for their unconditional support.
REFERENCES
Brookhuis, K.A. & de Waard, D. (2000). Assessment of drivers workload: Performance and subjective
and physiological indexes. Stress, workload and fatigue, CRC press, pp. 321–332.
Cafiso, S., Di Graziano, A. & La Cava, G. (2005). Actual driving data analysis for design consistency
evaluation. Transportation Research Record, Journal of the Transportation Research Board, 1912,
19–30.
Fitzpatrick, K., Wooldridge, M.D., Tsimhoni, O., Collins, J.M., Green, P., Bauer, K.M., Parma, K.D.,
Koppa, R., Harwood, D.W., Anderson, I.B., Krammes, R.A. & Poggioli, B. (2000). “Alternative
design consistency rating methods for two-lane rural highways.” FHWA-RD-99-172 Washington D.C.:
Federal Highway of Administration.
Green, P., Lin, B. & Bagian, T. (1993). Driver workload as a function of road geometry: A pilot experi-
ment. GLCTTR 22-91/01 Michigan: University of Michigan Transportation Research Institute.
Helander, M. (1978). Applicability of drivers’ electrodermal response to the design of the traffic envi-
ronment. Journal of Applied Psychology, 63(4), 481–488.
Hulse, M.C., Dingus, T.A., Fischer, T. & Wierwille, W.W. (1989). The influence of roadway parameters
on driver perception of attentional demand. Advances in Industrial Ergonomics and Safety, 1, 451–456.
Kannellaidis, G. (1996). Human factors in highway geometric design. Journal of Transportation Engi-
neering, ASCE, 122(1), 59–66.
Krammes, R.A., Brackett, R.Q., Shaffer, M.A., Ottesen, J.L., Anderson, I.B., Fink, K.L., Collins,
K.M., Pendleton, O.J. & Messer, C.J. (1995). Horizontal alignment design consistency for rural two-
lane highways. FHWA-RD-94-034, Washington D.C., Federal Highway Administration.
Krammes, R.A., Rao, K.S. & Oh, H. (1995). Highway geometric design consistency evaluation software.
Transportation Research Record, Journal of the Transportation Research Board 1500, 19–24.
Leisch, J.E. & Leisch, J.P. (1977). New concept in design speed applications, as a control in achieving
consistent highway design. Transportation Research Record, Journal of the Transportation Research
Board, 631, 4–14.
Messer, C.J. (1980). Methodology for evaluating geometric design consistency. Facility design and oper-
ational effects. Transportation Research Record, Journal of the Transportation Research Board, 757,
7–14.
Messer, C.J., Mounce, J.M. & Brackett, R.Q. (1981). Highway geometric design consistency related to
driver expectancy. FHWA-RD-81-035. Washington DC: Federal Highway Administration.
Mulder, G. (1980). The heart rate of mental effort (Doctoral thesis). University of Groningen, Gronin-
gen, Netherlands.
Stern, J.A., Walrath, L.C. & Goldstein, R. (1984). The endogenous eyeblink. Psychophysiology, 21,
22–33.
Suh, W., Park, P., Park, C. & Chon, K. (2006). Relationship between speed, lateral placement, and driv-
ers’ eye movement at two-lane rural highways. Journal of Transportation Engineering, 132(8), 649–653.
Weller, G., Jorna, R Gatti, G. (2005). Road user behaviour model. Sixth framework programme,
RIPCORD-ISEREST Deliverables D8.
Wooldridge, M.D., Fitzpatrick, K., Koppa, R. & Bauer, K. (2000). Effects of horizontal curvature on
driver visual demand. Transportation Research Record, Journal of the Transportation Research Board,
1737, 71–77.
Yerkes, R.M. & Dodson, J.D. (1908). The relation of strength of stimulus to rapidity of habit-formation.
Journal of Comparative Neurology and Psychology, 18, 459–482.
156

WebGIS enabled route planning system for tourists—a case study
Cynthia Baby Daniel & V.S. Manju

College of Engineering, Trivandrum, India
ABSTRACT: Tourism planning refers to the integrated planning of attraction, service (e.g.,
accommodation, restaurants, shops, medical facilities, postal services etc.,) and transporta-
tion facilities. This paper discusses the application of Geographic Information System (GIS)
as a route planning system for Tourists in Thiruvananthapuram City. Every tourist who is
alien to the place faces the problem of improper planning. The tourists may not be aware of
the route and the proper time to visit the tourist locations. A properly planned trip saves time
and money. Optimal Route is the route with least cost ie., least time or least distance. Given
a number of tourist location and the services required to be visited by the traveller, this mod-
ule computes the best route within a minimum time or minimum distance depending on the
impedance chosen. If a tourist specifies their origin and the tourist spots they wish to visit a
day, the system recommends the best route to travel based on the preferred time of visit or the
visiting hours of the locations. A web based GIS portal was also developed which would help
tourists get information about the study area in a single portal. It also has added capabilities
to the existing WebGIS system developed by the tourism department.
1 INTRODUCTION
Tourism planning is associated with the locations and interrelations of tourism infrastructure and
hence, they have to be analyzed within a spatial context. Therefore, GIS can be regarded as the
perfect platform for tourism planning. GIS is a rapidly expanding field that enables the develop-
ment of tourism applications in combination with other media. GIS operates mainly on two ele-
ments: Spatial data and Attribute data. Spatial data refers to the geographic space occupied by the
point and attribute data refers to the associated information such as name, category etc. In this
context, spatial data refers to the location of different tourist locations and the facilities such as
shops, hotels, movie theatres, bus stops etc., around them. Attribute data would include the road
name, road type, and travel time through each segment. Suggesting the optimal route, by consid-
ering the time window of the preferred locations, is of great help to tourists, who are unfamiliar
with the city. Route planning, an important stage of transportation planning can be applied here.
WebGIS refers to Geographic Information Systems that use web technologies as a method of
communication between the elements of a GIS. WebGIS provides a perfect tool to access, dis-
seminate and visualize tourism data. Any information that can be displayed on a digital map can
be visualized using WebGIS. The implementation of internet based GIS will provide interactive
mapping and spatial analysis capabilities for enhancing public participation in decision making
process. Web GIS have the advantages of global accessibility, better cross-platform capability,
diverse applications Sharma (2016). In addition to this, the capabilities of internet based GIS
will make it possible to answer spatial queries using intelligent maps with integrated images, text,
tables, diagrams; and showing locations of hotels, tourist sites, points of interest and so forth.
The transition from desktop GIS to WebGIS help people as it allows them the ability to search
information that could meet their needs freely and easily Sandinska (2016).
Gill & Bharath (2013) explored the application of GIS based Network analysis for route
optimization of tourist places in Delhi city. The study determined the optimal route from the
tourist origin to destination including visiting time at each tourist destination. Route analysis
157

is based upon the impedance, which is cost attribute (time or length) of road network. Arora &
Pandey (2011) used network analyst for finding out the level of service of ATMs of different
banks in South West Delhi. In addition to this, the shortest route from an accident location
to hospitals have also been found out using Network Analyst extension of ArcGIS. Nair
(2011) developed a Web based portal using Open Source GIS (OS GIS) to provide tourist
with location based information such as tourist spots, emergency services, travel hubs, hotels,
shopping markets etc. Kawano et al. generalized the problem of “Multi-Agents Scheduling
and Routing Problem with Time Windows and Visiting Activities”.
Varsha et al. (2016) developed a web based tourism application for Warrangal district of Tel-
engana. Zerihun (2017) developed a WebGIS portal for tourism in Ethiopia using open source
softwares. The transition from desktopGIS to WebGIS help people because it allows them the
ability to search information that could meet their needs freely and easily (Sanndinska 2016).
2 STUDY AREA
The study was conducted at Thiruvananthapuram, the capital of Kerala, the southernmost
state in India, The map of the study area is shown in Figure 1. The study area is around
226 km2. With a road density of 18 km per sq. km area, nearly seven per cent of the urban
land is under transport use. The National Highway NH 66 passes through the city. The Main
Central Road (MC Road) which is an arterial State Highway in Kerala and designated as
SH1starts from NH-66 at Kesavadasapuram in the city.
Trivandrum, with its pristine sandy beaches, long stretches of palm fringed shorelines, breezy
backwaters, historic monuments & rich cultural heritage embraces tourists of all kinds. Thiru-
vananthapuram is also a holy abode with many temples known for their excellent architecture &
hence a thriving pilgrim destination too. The major tourist attractions within the city such as
Aakulam and Veli tourist village, Napiers museum, Priyadarsini Planetorium, Kuthiramalika,
Kanakkakunnu palace, Sree Padmanabha Temple etc., are also shown in Figure 1.
3 METHODOLOGY
The step by step methodology adopted in this study is discussed briefly below. The analysis
was carried using Network Analyst of ArcGIS 10.1.
3.1 Data collection

Spatial data were collected from various Government Departments as well as Google Maps.
Road network was digitized from the base map obtained. Attribute data were collected from
Figure 1. Study area.
158

magazines, reports, by contacting the tourist locations and from various online tourism sites.
GPS survey as well as field survey was carried out to identify the tourist locations. Data
related to delay at roads were obtained from National Transportation Planning and Research
Centre (NATPAC).
The analyses were carried out using trial version of ArcGIS 10.1 downloaded from Euro-
pean Space Research Institute (ESRI) website. ArcGIS 10.1 offers many enhanced features
when compared to the previous versions of ArcGIS. Map source software was used to down-
load the GPS data. Garmin eTrex Vista HCx handheld GPS was used for GPS survey.
3.2 Creation of thematic layers

Different point layers that denote the facilities available around the places of interest such
as shops, hospitals, call taxis, movie theatres, hotels, bus stops, police stations, ambulance
services etc., were mapped. The location details were obtained using google maps as well as
GPS survey. The relevant attributes were added to these point layers and a spatial database
was created. Thus, an information system to help the tourists find the different facilities in the
study area forms the base module of this study.
3.3 Preparation of road network

The most important step in the analysis is the creation of the road network. The road network
map was created by digitizing the road map image obtained from Information Kerala Mission
(IKM). The roads were named and classified into three classes ie., National Highways, State
Highways and Major District Roads and assigned hierarchy values as 1, 2, 3 respectively. The
hierarchy values denote the ease of travelling through the roads. Network analyst uses it to cre-
ate more realistic routing based on the fact that people prefer to travel on larger roads. It is an
integer value that ranges from 1 (major) to 3 (minor) based on road class. One ways were identi-
fied and marked in the attribute table. Speed limit of heavy motor vehicles for Kerala was taken
as per the law by Kerala Motor Vehicles Department. For roads near schools, it is prescribed as
15 kmph. Hence buffers of 500 m were drawn around schools and those road links falling within
them were given a speed restriction. The travel time of each road segment was computed using
the specified speed and the length of the segment. The delay in each link due to various reasons
such as signals, junctions, bus stops, congestion etc., were obtained from the study conducted by
the National Transportation Planning and Research Centre (NATPAC). The final travel time of
each link was computed by adding the delay to the previously found travel time to give a more
realistic travel time. Bus is the mode of transport considered in the study.
3.4 Creation of network dataset

Network datasets are used to model transportation networks in ArcGIS. They were created
from source features, which include simple features (lines and points) and turns, and store the
connectivity of the source features. The road network created as a feature class is first con-
verted into topology to check for topological errors and then to network dataset so as to be
used in Network Analyst. Once the dangles and pseudo were corrected and the road network
was made free from errors, the Network Dataset was modeled.
3.5 Development of WebGIS module

A web based information portal was developed for the study area which was equipped with
the basic routing options. Once the GIS layers are created, the shapefiles are exported to a
spatial database in tabular form. The prototype adopted in this study is based on a three tier
architecture with three layers namely the client layer, database layer and application layer as
shown in Figure 2. The application layer consists of the web server and the GIS server. A
similar web structure was adopted by Kumar & Diwakar (2015) in developing a web based
159

Figure 2. Web architecture.
information system for Bhopal City, India. The clients access the portal through internet
browser through HTTP requests. The Web server is the software that responds to requests
from the client via the HTTP protocol. The Web server or the map server is a type application
server with manageability, processing and visualization of spatial data. The main purpose of
Map servers is the acquisition of spatial data from a spatial database. Tyagi (2014).
Geoserver has been used as the GIS server and Apache Tomcat acts as the Web server.
GeoServer is a Java-based software server that allows users to view and edit geospatial
data. GeoServer allows the display of spatial information to the world. Implementing the Web
Map Service (WMS) standard, GeoServer can create maps in a variety of output formats.
OpenLayers, a free mapping library, is integrated into GeoServer, making map generation
quick and easy. Apache Tomcat is an open source software implementation of the Java Servlet
and JavaServer Pages technologies. PostGIS is a spatial database. PostgreSQL is a powerful,
object-relational database management system (ORDBMS). PostGIS turns the PostgreSQL
Database Management System into a spatial database by adding support for the three features:
spatial types, indexes, and functions. As it is built on PostgreSQL, PostGIS automatically inher-
its important features as well as open standards for implementation. pgRouting is an extension
of PostGIS and adds routing functionality to PostGIS/PostgreSQL. pgRouting is a further
development of pgDijkstra. Dijkstra algorithm was implemented in the route finding module
of the WebGIS portal. Java was the scripting language used in both server and client side.
4 ANALYSIS AND RESULTS
The main analysis carried out in this study is optimum route analysis and route planning. This
section focuses on finding out the optimal route between tourist destinations. Optimal route is
that which has the least cost i.e., least distance or the least travel time. Given a number of tourist
location and the facilities required to be visited by the traveller, this module computed the best
route within a minimum time or minimum distance. Solving a route analysis can mean finding
the quickest or shortest, depending on the impedance chosen to solve for. If the impedance is
time, then the best route is the quickest route. Hence, the best route can be defined as the route
that has the lowest impedance, or least cost, where the impedance is chosen by the user. Any
cost attribute can be used as the impedance when determining the best route. The optimal route
between the selected locations is shown in Figure 3. The direction map is displayed in Figure 4.
The main component of this system is the route planning module which helps the tourists to
find the best sequence of visit to the desired tourist spots based on the time of visit of the tourist
destinations. This helps the tourists to optimise their trips and thus plan their day with minimum
waste of time. This would be of great help to the tourists, who are new to the place. Using the
identify tool, it is possible to obtain the image and other attributes of the desired locations. The
time of visit of the places along with its type will be displayed. In addition to this, it provides
the contact numbers as well as the holidays of the tourist place. If tourists specify the starting
location of their journey and the final destination from where they have to board, the sequence
of visit of the location is obtained based on the visiting time of the selected tourist locations.
The user has the option of preserving the first and last stops and specifying the starting time of
the journey as shown in Figure 5. The Route solver in Network Analyst generates the optimal
160

Figure 3. Sample optimal route.
Figure 4. Direction window.
Figure 5. Layer settings.
161

Figure 6. Sequence of visit of locations chosen by the user.
Figure 7. Home page of Web portal.
Figure 8. Sample optimal route and tourist location details.
sequence of visiting the stop locations as in Figure 6. The best itinerary based on the visiting
time of the selected tourist location chosen by the user is provided.
The home page of the interface is shown in Figure 7. This page welcomes the user into the
route planning system. Once the user clicks the enter button, the user enters into the next
page in which both the information system interface and the optimal route can be accessed.
The interface provides basic information about the tourist spots. On clicking the tourist spot
the details such as category, time of visit, contact number as well as the holidays are listed along
with the picture of the tourist spot as shown in Figure 8. Other layers such as bus stops, hospi-
tals, police stations, railway stations, movie theatres, shopping centres etc., are also displayed on
the map. By checking the required layers, it is possible to display them on the map. On clicking
on the specific points, their name can be viewed. On clicking the road network, the names of
the road can also be seen. The legend is also provided below. The zoom level can be adjusted as
per the need of the user and it is also possible to pan the map to focus the required area.
162

The routing option provided on the left side allows the user to find the optimal route
between any two points of interest. There are two options namely source type and destina-
tion type and a drop down menu associated with each of them. Once the type is set, the
user can choose the points of interest and on clicking the submit button, the optimal route
between the two points is displayed on the map window. Figure 8 displays the optimal route
between a tourist location and a railway station selected. The optimal route is computed
based on the travel time mentioned previously.
5 CONCLUSION
The application of GIS in tourist route planning is discussed in this study. The optimal route
option displays the route with shortest route or quickest route based on the impedence selected.
Route planning allows the tourist to plan his trip in an orderly manner without wasting time.
A WebGIS based portal was developed in this system which would help tourists to get infor-
mation about the location easily. The portal was developed with only the basic routing facili-
ties. This can be extended by including the other route planning functionalities mentioned in
this paper. The portal has been hosted in a local server and can be published in a public server
that can be accessed by all. Bus is the transport mode chosen for the study; the bus schedule
can be included at the implementation stage. The study can be further extended to incorporate
other modes of transport. An option for choosing the mode of transport can be provided.
ACKNOWLEDGEMENT
The study was funded by Transportation Engineering Research Center (TRC). The authors
would like to acknowledge National Transportation Planning and Research Centre (NAT-
PAC) and Information Kerala Mission (IKM). The authors would also thank Miraglo Pvt.
Ltd., for the technical help extended for developing the web portal.
REFERENCES
Arora, A. & Pandey, M. K. 2011. Transportation Network Model and Network Analysis of Road Net-
works, 12th ESRI India User Conference, December 7–8, Noida, India, 1–9.
Gill, N. & Bharath, B. D. 2013, Identification of Optimum path for Tourist Places using GIS based
Network Analysis: A Case Study of New Delhi, International Journal of Advancement in Remote
Sensing, GIS and Geography 1(2): 34–38.
Kawano, H. & Kokai, M. 2009. Multi-Agents Scheduling and Routing Problem with Time Windows
and Visiting Activities, The Eighth International Symposium on Operations Research and Its Applica-
tions, Sept. 20–22, 442–447.
Kumar, A. & Diwakar, P. S. 2015. Web GIS based Land information System for Bhopal City using
open Source Software and Libraries, International Journal of Science, Engineering and Technology
Research 4(1): 154–160.
Nair, S. 2011. Web Enabled Open Source GIS Based Tourist Information System for Bhopal City, Inter-
national Journal of Engineering Science and Technology 3: 1457–1466.
Sandinska, Y. 2016. Technological Principles and Mapping Applications of Web GIS, Proceedings, 6th
International Conference on Cartography and GIS, Albena, Bulgaria, 13–17.
Sharma N. 2016. Development of Web-Based Geographic Information System (GIS) for Promoting Tour-
ism in Sivasagar District, International Journal of Innovation and Scientific Research 24(1): 144–160.
Tyagi, N. 2014. Web GIS application for customized tourist information system for Eastern U. P., India,
Indian Society of Geomatics, 8(1): 1–6.
Varsha P. B., Venkata Reddy K. & Navatha Y. 2016. Development of Webgis Based Application for
Tourism, Esri India Regional User Conference – 2016.
Zerihun, M. E. 2017. Web Based GIS for Tourism Development Using Effective Free and Open Source
Software Case Study: Gondor Town and its Surrounding Area, Ethiopia, Journal of Geographic
Information System, 9: 47–58.
163

Comparison of demographic features, urban form and

travel characteristics
R.S. Remjish & S. Shaheem

Traffic and Transportation Division, National Transportation Planning and Research Center,
Thiruvananthapuram, Kerala, India
ABSTRACT: Referring to recent history, it is easy to perceive that there is an expansion of

urban areas and a need for improved transport infrastructure all over the world. These events
have led to congested, unsafe and inefficient vehicle transport networks along urban roads.
The situation is worse in the cities of developing countries like India because of the faster
growth. The promotion of education, increase in urban population density and change in the
lifestyle of city dwellers has paved the way for immediate requirements in extensive expan-
sion of road infrastructures. The extent of transport system expansion depends on the urban
form, trend of traffic movements and population. In this paper, an attempt is made to arrive
at some conclusion regarding the relationship between the urban area, traffic characteristics
and transport infrastructure based on a study conducted in the three major cities in Kerala –
Kochi, Thiruvananthapuram and Calicut.
1 INTRODUCTION
Extensive worldwide urbanization has led to unsustainable growth of cities. The increased
number of vehicles on urban roads in shorter span has produced a need for rapid expan-
sion of transport facilities which in turn has resulted in inefficient transport systems. The
percentage increase in the urban population from 1980 to 2000 was 48%, which is a drastic
difference. The urbanization level has almost stabilized in developing countries and a rapid
on-going process in the developing countries like India (Asian and African countries). Cities
need to expand to accommodate rural immigrants due to the increase in the area of cities.
Considering the extra space and providing the necessary infrastructure for the smooth life of
the dwellers, without affecting the overall level of service of road infrastructure, is a challeng-
ing task for transport system authorities and policy makers.
Stead and Marshall (2001) stated that urban planning has an important role to play in
helping to achieve more sustainable travel patterns. They pointed out that transport supply
and parking, as well as the distribution of land use, is influenced by planning policies hav-
ing an impact on travel demand and commuters’ mode. Frank and Pivo (1994) evaluated
the relationships between employment density, population density, land-use mix and single-
occupant vehicle (SOV) usage and found that they were consistently negative for both work
and shopping trips. They also found that the relationships between employment density,
population density, land-use mix, and transit and walking were consistently positive for both
work trips and shopping trips.
Advani et al. (2015) studied the relationship between urban population characteristics and
travel characteristics of different Indian cities with varying sizes and population densities.
Their findings reveal that city size plays a vital role in trip rate as well as trip length. They also
found that an increase in city size resulted in an increase in motorized trips. Crane (2000),
after a thorough investigation, expressed the importance of the geographic scale of the city
in favoring motorized trips, and the relevance of location of residences inside the urban area.
165

The deterioration of the urban environment through pollution of air, water and land is
another serious concern. The consumer suffers from increased travel time and the urban envi-
ronment suffers from pollution. The city authorities suffer from an inefficient usage of the
transport system supply and are faced with the prospect of increasing investment in trans-
port systems (which in most cases go on increasing and widening the existing road network to
alleviate congestion). In view of the various studies and evidence, evaluating the relationship
between the various parameters like urban form, population, travel characteristics and avail-
able transport infrastructure found out to be an immediate requirement.
2 BACKGROUND OF STUDY
2.1 Study area

The study area selected includes three major small sized cities in Kerala—Kochi,
Thiruvananthapuram and Calicut; including some adjoining areas for each.
2.1.1 Greater Kochin Development Authority (GCDA)

The greater Kochi area, with its present population of 2.1 million and employment of 0.7
million has a travel demand of 2 million passenger trips per day, of which 0.24 million trips
are performed during peak hours. With a growing population and mega development plans
coming up for this port city, the travel demand is expected to grow steeply. The greater Kochi
region has been aspiring to be one of the globally competitive cities in the country, but unfor-
tunately its transportation inadequacies are a notable shortcoming. With inadequate public
transport services, passengers will shift to private modes, which is already evident from the
high vehicle ownership trends in the region. This will not only aggravate the congestion on
the city roads but will also increase the pollution level. Observations from previous studies
also indicate increasing congestion levels in the city combined with a decreasing modal share
of the public transport. While the local urban bodies have been making concentrated efforts
to address the deficiencies, these efforts need to be intensified, fast tracked and integrated, in
order to reduce the social and environmental impacts on the community at large. As per the
National Urban Transport Policy (NUTP), it is essential to focus on the mobility of people
besides the movement of vehicles.
2.1.2 Thiruvananthapuram Study Area (TSA)

The total study area has a population of 1.33 million distributed over an area 374 sq.km
and a population density of 3,558 persons per sq.km. The population density is found to
be higher in the city area. Based on per capita vehicle ownership in Thiruvananthapuram
district, the vehicle population of the Thiruvananthapuram study area has been worked out
as 0.388 million vehicles. The public/semi-public area in the corporation has increased con-
siderably as the city acts as an administrative headquarters of Kerala State. In 2012, the area
covered by public/semi-public land use was 27.86 sq.km, which was 12% of the total area of
the corporation.
2.1.3 Calicut Study Area (CSA)

Kozhikode, formerly known as Calicut, is the third largest city in Kerala State after Kochi
and Thiruvananthapuram and is located in the Kozhikode district. The population, as per
the 2011 census, is 0.816 million within an area of 178.34 sq. km. The total number of house-
holds in the study area was 180,612. The city roads are maintained by both Public works
department (PWD) and the city corporation. The road network of Kozhikode city takes a
“spine and spur” form within the Central business district (CBD), with NH 66 (17) as the
arterial corridor and other roads darting out in the city. The vehicular traffic in Kozhikode
during 2001–2012 almost doubled, after which there was a slight decrease of 2.2% in the
growth of motor vehicles. As per recent statistics, the vehicular population in the Kozhikode
district stood at 570,267.
166

3 APPROACH
This paper focuses on the following aspects of medium sized cities in Kerala:
1. The urban form aspects like land area and population density.
2. Travel characteristics like trip length, trip rate and mode share.
3. Demographic aspects like income, household size and vehicle ownership.
The secondary data used for the study were taken from the Comprehensive Mobility Plan
study reports prepared for Kochi, Trivandrum and Calicut.
4 ANALYSIS
4.1 City area and population density

The relation between the geographical area and population density of the selected cities has
been studied. Figure 1 shows the inverse relationship between area and population density.
4.2 City area, trip length and trip rate

The relationship between city area and trip rate is depicted in Figure 2, from which it is clear
that the city with the lower area has a higher trip rate. This may due to the increased number
of short trips because people tend to have more trips per day, making different trips for dif-
ferent purposes, as the time required to complete a single trip would be less.
The relationship between the average city area and average trip length shown in Figure 3
indicates that the average trip length increases with an increase in city area.
4.3 Population density, household size, trip length and trip rate
It was found that the relationship between population density and average trip length shown
in Figure 4, is directly proportional, in that as the density decreases, the trip length increases.
This is due to the wide distribution of households throughout the city area.
Figure 1. City area and population density.
Table 1. Population density, average trip length and average trip rate for the
selected study areas.
City area Average trip length Average trip rate

City (sq.km) (km) (trips/person)
Kochi 632 8.10 1.04

Trivandrum 373.96 7.55 1.06
Calicut 178.34 6.97 1.18
167

Figure 2. City area and average trip rate. Figure 3. City area and average trip length.
Figure 4. Population density and average trip Figure 5. Population density and average trip
length. rate.
Figure 6. Household size and average trip rate.
Similarly, when comparing density and trip rate, as shown in Figure 5, the relationship is
inversely proportional, in that as density increases, the trip rate increases. This is because the
households, shopping areas, offices and other institutions are closely spaced and easy access
to these places tends the people to have frequent trips.
Upon comparing average household size and average trip rate, as shown in Figure 6, it was
found that the average trip rate increases with an increase in average household size.
4.4 City area and modal split

Assessing the influence of geographic features on travel characteristics is of a greater interest.
It allows the transport planner to create new strategies for sustainable urban behavior.
Table 2 shows the mode share of the three different cities. It was found that the city area does
not have much influence on mode share. This may be because the number of motorized trips
and mode share depend mainly on other factors like income, population and change in urban
168

Table 2. Mode share of the three cities.
Public Personal IPT NMT

vehicles (%) vehicles (%) (%) (%)
Kochi 42 36 7 15
Trivandrum 43 46 4 7
Calicut 41 34 7 18
Figure 7. Average income and vehicle ownership.
lifestyle than the city area. Even though there is a reasonable mode share for public transport,
the private vehicle share constitutes nearly half of the total mode share in Trivandrum city.
4.5 Per capita income and vehicle ownership

Another finding, shown in Figure 7, illustrates the relation between average income and vehi-
cle ownership. As the average income increases people tend to use more personalized modes.
The increase in per capita income might have modified the lifestyle of people, resulting in
urban inhabitants seeking more privacy and comfort.
5 CONCLUSION
Based on the above analysis, it was found that urban form, demography and travel charac-
teristics are highly interrelated in medium sized cities in developing countries. Understand-
ing the extent of the relationship among these parameters enables the transport planner to
forecast solutions for urban travel problems.
The following are the major conclusions derived from the study:
1. The area of the city and its population density has an inverse relationship between them.
2. As the city area increases, the average trip length and average trip rate increase.
3. It was found that the average trip length increases with a decrease in population density
and the average trip rate increases with an increase in population density.
4. The average trip rate increases with an increase in average household size.
5. Vehicle ownership increases with an increase in average income.
There is a need for change in urban lifestyles so as to maintain a balance between the
use of motorized modes and non-motorized modes such as walking, cycling and so on. It
was also noted that the share of private vehicles like two-wheelers and cars is increasing.
Providing good quality public transport networks may encourage users to shift to public
transport.
The study is limited to three medium sized cities in Kerala. Therefore, the relationship
between the considered parameters may vary depending on the location of the city, economic
169

growth rate of the city, shape of the city and promotion of industrial activities in the city.
For medium sized cities, controlling the expansion of the urban area and number of personal
vehicles, providing safe, comfortable and affordable public transport is needed for the devel-
opment of a sustainable transport system.
REFERENCES
Dominic, S. & Stephen, M. (2001). The relationships between urban form and travel patterns. An inter-
national review and evaluation, European Journal of Transport and Infrastructure Research (EJTIR),
1(2), 113–141.
Lawrence, D.F. & Gary, P. (1994). Impacts of mixed use and density on utilization of three modes of
travel: Single occupant vehicle, transit and walking. Transportation Research Record, 1466, 44–52.
Mukti, A., Neelam, J.G., Purnima, P. & Durai, B.K. (2015). Inter-relationship between transport sys-
tem, safety and city sizes distribution. Institute of Town Planners India, Journal, 12(4), 51–62.
Pranati, D. (2006). Urbanization in India, regional and sub-regional dynamic, population process in
urban areas. European Population Conference 2006, The University of Liverpool, UK, 21–24 June.
Randall, C. (2000). The influence of urban form on travel: An interpretive review. Journal of Planning
Literature, 15(1). 1–23.
Report on Comprehensive Mobility Plan for Thiruvananthapuram City. (2016). National Transporta-
tion Planning and Research Centre. December 2016.
Report on Comprehensive Mobility Plan for Kozhikkode City. (2016). National Transportation Planning
and Research Centre. December 2016.
Report on Comprehensive Mobility Plan and Parking Master Plan for Greater Cochin Region. (2015).
Urban Mass Transit Company Limited & Kochi Metro Rail Limited, Interim Report. November 2015.
170

Integration of a multi-modal transit system for urban areas:

A case study of Cochin city
S. Sreelekshmi & S. Shaheem

Traffic and Transportation Division, National Transportation Planning and Research Centre,
Thiruvananthapuram, Kerala, India
ABSTRACT: The integration of multimodal transportation systems explore the coordi-

nated use of two or more modes of transport for speedy, safe, pleasant and comfortable
movement of passengers in urban areas. It provides a convenient and economical connection
of various modes to make a complete journey from origin to destination. The present study
is aimed at reviewing various components for integrating multimodal transport systems and
in identifying various multimodal services such as metro, public transport, non-motorized
transport such as bicycles and para-transit services like autorickshaws in Kochi city. Data
regarding the public transport users were collected using a bus stop boarding and alight-
ing count survey as well as bus passenger interviews. Data regarding the socio-economic
and travel characteristics of the commuters were collected using a revealed survey. A stated
preference survey was also conducted by giving five attribute scenarios to the commuters and
their opinions were sought regarding their willingness to shift to a public transport system
varying in terms of speed, travel time as well as comfort and convenience.
1 INTRODUCTION
Cities in the developing world are in search of sustainable solutions to their accessibility
and mobility issues. The process is complicated by the rapid pace of urbanization, which is
characterized by motorization, the co-existence of motorized and non-motorized modes,
deteriorating public transport services and institutions, along with deteriorating air quality.
Travel demands have grown faster than the population and expansion of the city resulting in
delays for the movements to the city center/activity center from the suburbs/satellite towns
and vice versa. Public transit systems are struggling to compete with private modes in urban
cities and the shift is noticeable in developing countries as well, with; the predominant modes
being cars, two-wheelers and other intermediary modes. The agencies operating often fail to
respond to the demands. The resultant outcomes in most Indian cities have been increasing
congestion due to the increasing private modes, accidents and issues of rising air pollution
levels. The integration of multimodal transportation systems explores the coordinated use of
two or more modes of transport for speedy, safe, pleasant and comfortable movement of pas-
sengers in urban areas. It provides convenient and economical connections of various modes
to make a complete journey from origin to destination. Generally, it has been characterized
by increased capacity, efficient access, reduced road congestion, longer journey times and
reduced air pollution. The Integrated Multi Modal Transport System (IMMTS) comprises
of one trip that involves two or more than two different modes of transportation. Muley and
Prasad (2014) defines integration of multimodal transit as the way parts of public transport
network are embedded in total mobility chain. The study conducted by Cheryan and Sinha
(2015) identified the increasing need of travel due to rapid urbanization leading to increasing
congestion and necessity for the development of public transit system by providing an inte-
grated transit service. The major goal of integration of a multimodal transport system is to
promote public transport in urban areas. A well-coordinated integration of different modes
brings about greater convenience for commuters, efficiency and cost effectiveness.
171

2 SCOPE AND OBJECTIVES
2.1 Study area

The greater Kochi area has a population of 2.1 million and need to address a travel demand
of 2 million passenger trips per day, of which 0.24 million trips are performed during peak
hours. Cochin city is aspiring to be one of the globally competitive cities in the country with
one of the major limitations being transportation inadequacy. There is a major portion of
commuters shifting to private modes. This will not only aggravate the congestion on the city
roads but will also increase the pollution level. It is essential to focus on the mobility of peo-
ple besides the movement of vehicles as per the National Urban Transport Policy (NUTP).
The area for the present study is shown in Figure 1.
2.2 Scope and objectives

The scope of the study was limited to two major proposed metro stations: Edappally and
Kaloor in Kochi City. These two stations were selected as there was mass movement of peo-
ple across the influence zone of these two stations when compared to the others. The objec-
tives of the study were:
• To review the existing transport system of Kochi city
• To study different modes of transport such as bus, private vehicles like cars and two-wheelers
as well as para-transit services like auto-rickshaws
• To develop various strategies for multimodal integration
3 METHODOLOGY
The core part of the study lies in the formulation of a well-defined methodology that ensures
an effective proposal of suggestions for an efficient integration of multiple modes in the city
of Kochi. The methodology adopted for the study is shown in Figure 2. At first, a detailed
literature survey was conducted to investigate studies on the integration of multimodal sys-
tems in various cities. Then, the study areas were selected which were the two metro stations
at Kaloor and Edappally. Data were collected from the study areas by conducting different
surveys among the commuters using car, bus, two-wheeler and auto rickshaw as modes. The
surveys conducted were a boarding and alighting survey, bus passenger interviews, as well as
a stated preference survey. The collected data were analyzed and suitable recommendations
were framed.
Figure 1. Study area.
172

Figure 2. Methodology adopted for the study.
3.1 Data collection

The surveys conducted to collect relevant data regarding traffic volume and passenger
demand existing in the study areas were:
Bus Stop Boarding and Alighting Survey: This survey was conducted for 14 hours at the
major bus stops of the study areas covering morning and evening peak periods. The survey
format captured the time of arrival at the bus stop, route number/name, origin of bus, desti-
nation of bus, number of passengers boarding and alighting the bus and occupancy details.
Trained enumerators were stationed at identified bus stops capturing the requisite data.
Bus Passenger Interview Survey: The passengers waiting to board the buses and alighting
from the buses were interviewed to capture the passenger travel characteristics throughout the
study area. The surveys were carried out at the identified locations for one working day devoid
of any rainfall/festivities and so on, for a period of fourteen hours (06:00 am to 8:00 pm)
capturing the bus passengers on either side of the road, best representing the regular com-
muting pattern of the passengers. Data corresponding to socio economic as well as travel
characteristics of the commuters were collected. The major socio economic characteristics
collected included age, gender, monthly income, education, occupation and monthly expenses
on transport.
Stated Preference Survey (SP Survey): The SP survey questionnaire was designed to iden-
tify the travel patterns and preferences of the interviewed respondents. The first section
of the questionnaire was designed to ask questions covering respondents’| existing travel
patterns and socio-economic characteristics. The second section of the questionnaire was
related to the perception of the existing public transport system in the study area. The last
section on both the questionnaires was supported by five SP survey travel attribute scenarios.
It was intended to provide the respondents with varying scenarios with existing and proposed
system attributes such as travel time, waiting time, fare/cost and comfort level.
3.2 Data analysis

a. Bus Stop Boarding and Alighting: The survey helped in evaluating bus demand in each
time interval. It gave the peak number of commuters boarding and alighting at both of the
173

stations and also aided in arriving at the peak hour of demand for buses. At Kaloor, it was
observed that the evening peak was from 5:00 pm to 6:00 pm and morning peak time was
from 9:00 am to 10:00 am for boarding passengers, whereas the evening peak was from
4:00 pm to 5:00 pm and morning peak time was from 9:00 am to 10:00 am for alighting
passengers. At Edapally, it was found that the evening peak was from 5:00 pm to 6:00 pm
and morning peak time was from 8:00 am to 9:00 am for boarding passengers, whereas
the evening peak was from 6:00 pm to 7:00 pm and morning peak time was from 8:00 am
to 9:00 am for alighting passengers. Table 1 shows the peak traffic volume from each bus
stop of the study area.
b. Bus Occupancy Survey: Along with the bus stop survey, the bus occupancy survey was also
carried out. Higher proportions of occupancy at each of the stations during both peak
hours and off-peak hours are as shown in Table 2.
c. Bus Passenger Interview Survey: Data regarding socio-economic characteristics of the
travelers were collected. The following details of the passengers traveling by bus regarding
their access as well as egress trips were also obtained as part of the survey:
Access and Egress Modes: Share of access mode from origin to boarding point and dis-
persal mode from alighting point to destination of the boarding bus passengers are shown
in Figures 3 and 4. It was found that bus is the predominant mode of travel from origin to
boarding point while walking is the predominant mode from alighting point to destination
of the bus users.
Proximity to the Bus Stops: This showed the closeness of bus stops either to the point of
origin or destination. To analyze the proximity of bus stops, the distance from the origin to
the bus stop and from the bus stop to the destination were considered. Figure 5 shows the
distribution of access and egress trip distances.
Table 1. Peak hour volume of boarding and alighting passengers.
Peak volume of Peak volume of

Location boarding passengers alighting passengers
Edappally 1412 821

Kaloor 577 1116
Table 2. Occupancy level.
Location Peak hour occupancy level Off-peak hour occupancy level
Edappally 45% of the buses have an occupancy 30% of the buses are full
level of full sitting plus full standing sitting
Kaloor 40% of the buses are full sitting plus 80% of the buses are half
half standing sitting
Figure 3. Access mode from origin to boarding Figure 4. Egress mode from alighting point to
point. destination.
174

Figure 5. Proximity to the bus stops for bus Figure 6. Waiting time at the bus stop.
passengers.
It was found that approximately 31% of the access trips to the bus stops were more than
5 km and 40% of the egress trips from the bus stops were within 500 m. Another key obser-
vation is that more than 30% of the access and egress trips have a trip distance greater than
1 km.
Waiting Time at the Bus Stop: Figure 6 represents the distribution of commuters according
to the waiting time at the bus stop. It was found that about 46% of the boarding passengers
wait up to five minutes to board the bus followed by 42% of commuters having a waiting time
of 7.5 minutes to 15 minutes.
S P Survey: The major observations from the survey are as follows:
Mode and Trip Purpose: Figure 7 gives the distribution of trip purpose among commuters
considering different types of mode. The majority of the commuters travel with a trip pur-
pose of work followed by business and education.
Mode and Travel Time: Figure 8 gives the distribution of travel time among respondents
considering different types of mode. It was found that the majority of the commuters have a
travel time of 20–40 minutes. It was also found that the majority of the commuters traveling
by bus have a travel time greater than 40 minutes.
Mode and Travel Cost: Figure 9 gives the distribution of travel cost among respondents
considering different types of mode. It was found that the majority of trips involving a car
have travel costs greater than 40 Rs and trips using autorickshaws have travel costs less than
20 Rs.
Mode and Travel Distance: Figure 10 represents the distribution of trip length by different
modes of transport. It was found that the trips having a travel distance greater than 20 km
are mostly performed by bus. It was also found that the commuters having a travel distance
less than 20 km prefer cars and two-wheelers as their modes.
Monthly Income and Mode of Travel: The main purpose of comparing monthly income
with mode of travel is to observe the category of commuters with different income levels
using different types of mode. It was found that the higher income commuter groups mainly
prefer cars and two-wheelers. Figure 11 represents a comparison of different modes of travel
with income. It was found that the higher income groups did not prefer public transport,
whereas low income groups prefer public transport for their travel.
Daily Travel Cost and Monthly Income: Figure 12 represents the comparison of monthly
income with travel cost.
Stated Preference: A S P survey was also conducted by giving five different scenarios
to the passengers and their opinions were sought regarding their willingness to shift to a
public transport system varying in terms of speed, travel time, as well as comfort and con-
venience. The details of each scenario with its characteristics are shown in Table 3. It was
found that about 49% of the users were satisfied with the existing scenario. About 27% of
the commuters were willing to shift to public transport if travel time and waiting time are
reduced by 25%.
175

Figure 7. Comparison of mode with trip Figure 8. Comparison of mode with travel time.
purpose.
Figure 9. Distribution of travel costs by mode. Figure 10. Comparison of mode with travel
distance.
Figure 11. Comparison of mode with income.

Figure 12. Comparison of monthly income
with daily travel cost.
Table 3. Choice set for public transport users.
Scenarios
Parameters 1 2 3 4 5
Waiting time Existing Existing Reduced by 25% Reduced by 25% Reduced by 50%
(min)
Travel time Existing Existing Same as existing Reduced by 25% Reduced by 25%
(min)
Travel cost Existing Existing Increased by 25% Increased by 25% Increased by 50%
(Rs.)
Transfers Existing Yes Yes No No
Comfort Existing Crowded + Sitting + Standing + Sitting + Non-AC Sitting + AC
level Non-AC Non-AC
Percentage of 49% 11% 10% 27% 3%
users opting
176

4 INTEGRATION STRATEGIES
The major integration strategies adopted for the study are the following:
a. Physical Integration
1. W alkability: Walking was found to be the predominant mode from alighting point to
destination of the bus users. Pedestrian facilities need to be enhanced and the key fac-
tors to making walking appealing are safety, activity and comfort. For comfortable and
safe walking, footpaths of a minimum width of 1.8 m should be provided.
2. Cyclability: Cycling is an emission free, healthy and affordable transport option that is
highly efficient and consumes little space and few resources. It combines the conven-
ience of door to door travel, the route and schedule flexibility of walking and the range
and speed of many local transit services.
3. Connectivity: Short and direct pedestrian and cycling routes require highly connected
networks of paths and streets around small, permeable blocks. A tight network of
paths and streets offering multiple routes to many destinations can also make walking
and cycling trips varied and enjoyable. Frequent street corners and narrower rights of
way with slow vehicular speed and many pedestrians encourage street activity and local
commerce.
4. Multimodal Integration: Integration of mass transit with the city bus system and
extended to the Non Motorized Modes such as bicycles as well as Intermediate Para
transit system not only ensure first and last mile connectivity but also the seamless
transfer of network users. Multimodal integration exists at various stages such as: insti-
tutional, physical, fare, operational. The level at which multiple modes need to be inte-
grated shall depend on the local city authorities.
5. Information Integration: This is the information system placed on-board in the bus. It is
a real time information display unit which helps in providing passengers with the neces-
sary information related to their commute. The information displayed varies from the
next/current stop information, current location, expected time to reach the destination
or the nearest metro station and so on. Route data may be presented as a linear map,
highlighting the current position of the bus and the next stop that it is approaching.
These maps are held as image files on the vehicle computer and displayed when the bus
reaches a pre-specified geolocation. Likewise, audio files can be delivered in exactly the
same way, notifying passengers of a particular stop as the vehicle approaches.
b. Fare Integration
In the current scenario, it is imperative that the fare collection system for the city bus trans-
port, feeder and metro be integrated for the ease of the passengers. The data from ticket
issuing machines should be downloaded into the central room server for further analysis.
These data which include denomination wise ticket sales, number of passengers, load factor
at each stage of the route and so on, are stored in the computer for further analysis. Based
on these data, the operations of the bus system can be monitored and documented for
future planning of operations. These data will be of great help for route restructuring, route
analysis, introduction/curtailment of routes and so on. Two important components of fare
integration are:
1. Hand held ticketing devices are small ticket vending units which are capable of producing
tickets as per the specified distance or stoppages.
2. Smart cards are small plastic cards which hold a certain value of money and can be used
as a fare paying system at metro stations.
5 CONCLUSIONS
The existing transport system of Kochi city was analyzed as part of the study. At Edappally
and Kaloor, the majority of the public transport buses had an occupancy level exceeding their
177

capacity. During peak hours, about 45% of the buses had an occupancy level of full sitting plus
full standing at Edappally, whereas the occupancy level was full sitting plus half standing at
Kaloor. The results of the S P survey indicate that the majority of the commuters, which consti-
tuted about 49% of the total, were satisfied with the existing transport conditions. About 27%
of the commuters were willing to shift to a public transportation system with travel attributes
such as travel time and waiting time being reduced by 25%, inspite of the increase in travel costs.
Moreover, walking was found to be the predominant mode from alighting point to destination
of the bus users. Thus, pedestrian facilities need to be enhanced and the key factors to making
walking appealing are safety, activity and comfort. Significant integration strategies were also
recommended as a part of the study involving both fare integration using hand held ticketing
devices and smart cards, as well as physical integration strategies involving walking, cycling and
connectivity.
REFERENCES
Arentze, T.A. & E.J.E. Molin (2013), Travelers’ preference in multimodal networks: Design and results
of a comprehensive series of choice experiments, Transportation Research Part A 58 (2013), 15–28.
Cheryan, C.A. & S. Sinha (2015). Assessment of transit transfer experience: Case of Bangalore, 8th
Urban Mobility India Conference & Expo.
Li, L., J. Xiong, A. Chen, S. Zhao, & Z. Dong (2014). Key strategies for improving public transportation
based on planned behaviour theory: Case study in Shanghai, China, Journal of Urban Planning and
Development 141(2): 04014019.
Muley, B.R. & C.S.R.K. Prasad (2014). Integration of public transportation systems. Urban Mobility
India Conference & Expo, Transportation Division NIT Warangal.
Najeeb, P.M.M. (2008). Study of cognitive behavior therapy for drivers improvement, Australasian
Road Safety Research, Policing and Education Conference, Adelaide, South Australia.
Patni, S. & Ghuge, V.V. (2015). Towards achieving multimodal integration of transportation systems for
seamless movement of passengers: Case study of Hyderabad City. Urban Mobility India Conference
& Expo.
178

Quantitative evaluation of bus terminal using time-space analysis
V. Hridya & George Geeva

Department of Civil Engineering, Rajiv Gandhi Institute of Technology, Kottayam, Kerala, India
ABSTRACT: Bus terminals are enclosures where the interactions among passengers are
very high. Activities like walking, waiting, route choice, boarding and alighting of buses
occur. Pedestrian level of service is an overall measure of walking conditions on a path, facil-
ity or route. In this study, the quantitative evaluation of a terminal is conducted using time
space analysis. The selected site for the study is Aluva Kerala state road transport corpora-
tion bus terminal. The main aim of the study is to determine the level of service of the termi-
nal. Queuing and walking are considered as the major passenger activities inside a terminal.
Different activities require a different amount of space and time. The time and space utilized
by passengers for standing and circulation is considered.
1 INTRODUCTION
Transportation is defined as a system of interacting components or modules that are speci-

fied as passenger processing elements which aid pedestrian movement, and the environment
which encompasses those dimensions with which the pedestrian associates his or her per-
sonal comfort, convenience, safety and security [Demestsky, M.J., Hoel, L.A. & Virkler, R.M.
1977]. The combined performance of subsystems for passenger processing and environment
conditions accounts for the overall effectiveness of station design [Demestsky, M.J., Hoel,
L.A. & Virkler, R.M. 1977].
A bus terminal is the point of entry or exit to the public transportation system. It may also
act like a point for concentration, dispersion, and interchange of modes, storage space for
passengers and vehicles, as well as a maintenance area. A terminal that accommodates a high
passenger volume can be looked upon as two interacting components—passenger process-
ing elements and environmental conditions. The effective interactions of these components
control the performance of the bus terminal.
The Highway Capacity Manual’s (HCM) methods for analyzing pedestrian Level of Serv-
ice (LOS) are based on the measurement of pedestrian flow rate and sidewalk space. The
pedestrian flow rate, which incorporates pedestrian speed, density and volume, is equiva-
lent to vehicular flow. Pedestrian speed as observed in HCM 2000 as follows: “As volume
and density increase, pedestrian speed declines. As density increases and pedestrian space
decreases, the degree of mobility afforded to the individual pedestrian declines, as does the
average speed of the pedestrian stream.”
1.1 Need for study

The cities and nations who had invested more in efficient transportation systems are now
very weak in their environmental sustainability. A terminal is a point where the bus starts or
ends its schedule and hence it makes a clear contribution to the overall journey experience
of the passenger. The passenger may have some expectations about the terminal, about the
access of the terminal, frequency of the buses, the services and the facilities provided and so
on. If these criteria do not meet the expectations of the passenger, this may result in a shifting
of travel mode from public to private.
179

The relevance of shifting from private to public vehicles is increasing nowadays. A study
on the quality of facilities provided at the terminal will help to gain ideas regarding the
improvements to be done by the authority.
1.2 Literature review

1.2.1 Different types of users
The different types of passengers using a transit platform are identified as commuters/busi-
ness travelers, leisure travelers, passengers in wheelchairs, those with a physical or cognitive
mobility impairment, people with medium sized luggage, people with large sized luggage and
parents with small children [Department of New York Rails, 2011]. Passengers can also be
categorized as general users and special users. The physically handicapped and elderly are
considered as special users [Demestsky, M.J., Hoel, L.A. & Virkler, R.M. 1977].
1.2.2 Factors considered for evaluation

As a terminal is a place of integration of passengers, buses, other services and so on, the facilities
provided are of greater importance. Various factors identified are physical characteristics, loca-
tion facilities and user factors as per the Guidelines for Assessing Pedestrian Level of Service-
(Main roads Western Australia, 2006). Suitability for use, path width, surface quality, crossing
opportunities and support facilities are the physical characteristics. Location facilities include
connectivity, path environment and potential for vehicular conflict. User factors selected are
pedestrian volume, mix of path users and personal security. The physical environment includes
lighting, air quality, temperature, aesthetics advertising, cleanliness, music and so on. Non-trans-
port businesses and services, private concessions such as newspaper stands, coffee shops, barber
shops and other small businesses and services, restrooms and lounges, first aid stations, public
telephones, weather protection facilities can also be considered as performance evaluators in a
bus terminal [Demestsky, M.J., Hoel, L.A. & Virkler, R.M. 1977]. Safety, amenities, efficient
movement are the other factors described in the Public Transport Infrastructure Manual, 2016.
1.2.3 Time space analysis

In qualitative analysis the time and space utilized by passengers is considered. Time space
analysis is a method of analysis used to determine the pedestrian traffic in urban streets
which exhibits the need for holding space and circulation space. Pedestrian LOS is an overall
measure of walking conditions on a path, facility or route. This is directly linked to factors
that affect pedestrian mobility, comfort and safety. In a study by Grigoriadou and Braaksma
(1986), the pedestrian activities like queuing and circulation were taken into consideration. In
platforms there would be holding space for standing and circulation space for walking. The
time and space occupied by the passenger activities are incorporated to determine the space
requirement of pedestrians. The passenger LOS of terminal areas are developed based on
average pedestrian space, personal comfort and degrees of internal mobility. The standards
are presented in terms of average area per person and average interpersonal space.
2 STUDY AREA
Aluva formerly called as Alwaye, is the second biggest town of Greater Cochin City in the
Ernakulam district of Kerala. It is also considered as the industrial and commercial city of
Kochi and it is the industrial epicenter of the state. Aluva is located on the banks of River
Periyar. Aluva is also a major transportation hub, with easy access to all major forms of
transportation. Aluva acts as a corridor which links the highland districts to the rest of the
state. Aluva is more famous for its accessibility through rail, air, metro and by bus.
Aluva KSRTC bus terminal is one of the major bus depos in Kerala. Buses from all other
parts of Kerala have services to this station. Bus services from places like Mysore, Mangalore,
Bangalore, Trichy, Coimbatore, Salem, Palani, Kodaikanal and so on are also available. At
present, the depo has 107 buses of its own and seven interstate services.
180

Figure 1. Layout of Aluva KSRTC bus terminal.
After the pilot survey of different bus stations, it was noticed that Aluva KSRTC bus termi-
nal is the one with less facilities compared to the other stations nearby. The study of this bus
terminal was very timely in order to determine the pedestrian activities within it and thereby,
to determine the LOS. Aluva KSRTC bus terminal acts as a center that connects different
parts of the state. Because of that, many categories of passengers use the bus service from
this station for different purposes such as work trips, business trips, shopping and recreational
trips. The major share of the passengers includes students and private sector officials.
Figure 1 shows the layout of Aluva KSRTC bus terminal. The entry and exit points for
buses and passengers are shown and also the different zones selected for study are depicted
in the layout.
3 METHODOLOGY
3.1 Quantitative evaluation using time-space analysis

Bus terminals are critical nodes in a bus transit network. “Passengers” safety, comfort, con-
venience and so on will be affected by the platform activities. Pedestrian activities like queu-
ing and circulation that is standing and walking, are mainly occur on a terminal platform.
The walking and circulation presents different time as well as space characteristics. The time-
space method works on the principle that the walking pedestrians and the standing pedestri-
ans will use a different amount of space for different time periods. This method of evaluation
is a type of quantitative evaluation since the quantity of time and space available in the bus
terminal are considered. Table 1 shows the LOS for different pedestrian modules for a queu-
ing area [Transport Research Board, 2000].
The selection of the study site was very important since the waiting area was very con-
gested and limited without any proper planning. Therefore, it was necessary to determine
where all the passengers would wait and spend their time in between journeys. This would
help to identify how people utilize the available space.
3.2 Calculation steps

The calculations were done using the following equations [Grigoriadou, M. & Braaksma,
J.P. 1986].
i. Calculate the total available time space (TS)
TS = A × AP (1)
181

Table 1. LOS, pedestrian module and interpersonal space as per HCM.
LOS Module (m2/ped) Interpersonal space (m) Comments
A 1.2 or more 1.2 or more free circulation much interpersonal space

no crowding
B 0.9–1.2 1.1–1.2 very little crowding sufficient interpersonal
space
C 0.7–0.9 0.9–1.1 restricted circulation small interpersonal
space walking speed controlled
D 0.3–0.7 0.6–0.9 severely restricted circulation limited
interpersonal space no personal contact
E 0.2–0.3 0.6 or less standing room only little interpersonal space
no circulation possible
F 0.2 or less Touching standing only no interpersonal space jammed
conditions
where TS = total available time-space, in m2.min; A = visible platform area, in m2; and
AP = analysis period, in min.
ii. Calculate holding area time space requirements (TSh)
TSh = SP × AST × AASP (2)
where TSh = holding area time space requirements, in m2.sec; SP = number of standing pedes-
trians; AST = average standing time, in sec; AASP = standing pedestrian module, in m2/ped.
iii. Calculate the net circulation area time-space (TSc)
TSc = TS – (TSh/60) (3)
where TSc = net circulation area time-space, in m2.min; TS = total available time-space, in

m2.min; and TSh = holding area time space requirements, in m2.sec.
iv. Calculate the total circulation time (TC)
TC = CP × ACT (4)
where TC = total circulation time, in ped-min; CP = total number of circulating pedestrian;

and ACT = average circulation time, in min.
v. Calculate the pedestrian module for walking pedestrian (MOD)
MOD = (TSc/TC ) (5)
where MOD = pedestrian module for walking pedestrian, in m2/ped; TSc = net circulation

area time-space, in m2.min; and TC = total circulation time, in ped-min.
vi. Compare the calculated pedestrian module with the standard values shown in Table 1 to
determine LOS.
4 DATA USED
The study site was Aluva KSRTC bus terminal. The quantitative analysis of the terminal
would help us to identify the real conditions of the terminal. After conducting a pilot survey,
the different zones inside the terminal were identified. From the survey, five different zones
where major pedestrian activities occur were selected. The zones are the waiting area, area
near the bus bay, hotel or restaurant, mini snacks bar area and shopping area. Areas of all
the zones were measured manually. A pedestrian volume survey was conducted to obtain the
182

Table 2. Peak hour passenger count and average time spent by passengers at each zone.
Thursday Friday Saturday

Average
Name of No. of time
the zone pedestrians Morning Evening Morning Evening Morning Evening (min)
Area near Standing 192 228 296 272 116 200 12.25
bus bay pedestrians
Walking 224 164 272 312 168 188
pedestrians
Waiting Standing 228 256 372 300 240 184 10.8
area pedestrians
Walking 220 180 180 272 140 176
pedestrians
Shopping Standing 108 164 104 228 44 80 8.2
area pedestrians
Walking 76 80 64 140 88 88
pedestrians
Hotel or Standing 32 40 60 76 48 76 15.58
restaurant pedestrians
Walking 24 24 8 60 60 72
pedestrians
Mini snacks Standing 40 56 80 108 64 80 6.8
bar area pedestrians
Walking 68 60 68 100 96 92
pedestrians
number of passengers occupying each zone and the average time spend by passengers for
each activity. Standing and circulating activities of passengers are included in this study.
The survey was conducted manually for three different days, including a typical working
day, a weekend and a holiday. The survey was conducted during the morning peak hours and
evening peak hours in a time interval of 15 minutes. Table 2 shows the number of passengers
present in each zone during the time of the survey. The morning peak time was identified
from a pilot study from 8.00 am to 10.00 am and evening peak time was from 3.30 pm to
6.00 pm. Data in the table depicts the passengers in every hour of survey. The data collected
include the area of all the zones, number of standing pedestrians, number of circulating
pedestrians, average standing time of pedestrian, average circulation time of pedestrians and
standing pedestrian module. After the data collection, the calculations were done using the
equations given in section 3.2.
The calculation steps were automated using a tool called the Structured Query Language
developer. The SQL developer uses computer programming language called SQL. SQL is
a set of statements with which all programs and users access data in an Oracle database.
The SQL provides benefits to managers, end users, application programmers and database
administrators. Technically speaking, SQL is a data sublanguage. It works with a set of data
rather than individual units. For the data input we have to create a survey table which includes
all the data collected from the survey. Then, we have to prepare a query with equations used
for calculation for reading the data from the table and applying it to the calculation steps.
The final result is that the pedestrian module will compare with Table 1 and then the LOS is
determined.
The MOD calculated for each zone is in the range of 0.2 or less which indicates the LOS
grade is ‘F’. This indicates that the bus terminal has a poor performance with congested con-
ditions. There is interpersonal space and a stand only condition exists at peak hours.
183

6 CONCLUSION
• Walking and queuing were the major activities occurring in the terminal.
• The evaluation was done using time-space analysis.
• The time and space utilized by passengers were considered.
• Five zones of the terminal were identified where the major passenger activities occur.
• The area of all the zones, number of standing pedestrians, number of circulating pedes-
trians, average standing time of pedestrians, average circulation time of pedestrians and
standing pedestrian module were determined.
• In a quantitative aspect or by means of time-space analysis, the LOS was found as LOS
grade F, which indicates a congested and a standing only possible condition.
• The results indicate low quality of terminal and lack of space availability during peak
hours.
7 LIMITATIONS AND SCOPE
The data collection was done manually due to the complex structure of the bus terminal and
due to some administrative implications. This could be a videographic survey to gain more
precise values.
The work could be extended to greater number of terminals.
The comparison of the LOS values of different terminals could be done.
Only a quantitative evaluation was carried out in this study. Quantitative as well as qualita-
tive analysis could be incorporated in the same study.
Classification on the basis of long distance travelers, short distance travelers and so on
could be done.
ACKNOWLEDGMENTS
The authors gratefully acknowledge the effort of classmates of the final semester trans-
portation engineering programme at RIT, Kottayam, who played a great part in the data
collection.
REFERENCES
Chang, C.C. (2009). A model for the evaluation of airport service quality. Proceedings of the Institution
of Civil Engineers transport 162, 4, 207–213.
Correia, A.R. & Wirasinghe, S.C. (2008). Analysis of level of service at airport departure lounges: user
perception approach. Journal of Transportation Engineering, 134, 105–109.
Demestsky, M.J., Hoel, L.A. & Virkler, R.M. (1977). A procedural guide for the design of transit stations
and terminals. Department of civil engineering, University of Virginia, Charlottesvi, le, Virginia.
Department of NewYork Rails. (2011). Station capacity assessment guidance.
Eboli, L. & Mazzulla, G. (2007). Service quality attributes affecting customer satisfaction for bus tran-
sit. Journal of Public Transportation, 10 (3), 21−34.
Geeva, G. & Anjaneyulu, M.V.L.R. (2015). Development of quality of service index for bus terminals,
Proc. 2nd Conference on Transportation Systems Engineering and Management. NIT Tiruchirappalli,
India, May 1–2.
Grigoriadou, M. & Braaksma, J.P. (1986). Application of the time-space concept in analyzing metro
station platforms. Journal of Institute of Transportation Engineers, 33–37.
Litman, T. (2008). Valuing transit service quality improvements. Journal of Public Transportation, 11, 2.
Main Roads Western Australia. (2006). Guidelines for assessing pedestrian level of service. 1–9.
Transit Translink Authority. (2012). Public Transport Infrastructure Manual.
Transport Research Board. 2000. Highway capacity manual, National Research Council, Washington D.C.
184

Modeling of the transport of leachate contaminant in a landfill

site: A case study in Mangaluru
Anand Divya & S. Shrihari

Department of Civil Engineering, National Institute of Technology, Surathkal, Mangaluru, Karnataka
H. Ramesh
Department of Applied Mechanics, National Institute of Technology, Surathkal, Karnataka, India
ABSTRACT: Ground water flow and the solute transport model MODFLOW and
MT3DMS were established to determine the spread of contamination from a landfill main-
tained by Mangaluru City Corporation at Vamanjoor, located nearly 8.5 km from the center
of the city. As Vamanjoor is home for many educational institutes and also a residential area,
the spread of the contamination has to be analyzed. For this study, the aquifer considered is
a subbasin of the Gurupur basin. This study has focused on handling the data available in
the most efficient way to develop a consistent simulation model. The model was calibrated
successfully with RMSE value of observed versus simulated head as 0.32 m. The evaluation
of model was also done by comparing with the measured water head and chloride level from
the field on a seasonal basis. After validating successfully, the model was run to determine
the extent of contamination and also to forecast a scenario for maximum rainfall. The results
show that the contamination has spread to a distance of 1 km from the landfill and with
maximum rainfall the spread will be around 1.8 km from the landfill.
1 INTRODUCTION
Solid waste, when dumped into an uncontrolled landfill, can cause a serious threat to an under-
lying aquifer due to the migration of leachate generated from the landfill into the groundwater.
The amount and nature of waste dumped in a landfill depends upon various factors such as
inhabitants of the city, people’s lifestyle, food habits, standard of living, the degree of industri-
alization and commercialization of that area, culture and tradition of inhabitants, and also the
climate of the area. Due to unscientific collection, transportation and disposal of solid waste
without environmentally friendly methods such as composting, incineration and so on, dump-
ing of waste in India has become more chaotic. The leachate characteristics also depend upon
the pre-treatment of the solid waste such as separation of recyclable material like plastics, paper,
metals, glass and so on, as well as grinding or bailing of the waste (Kumar & Alappat, 2005).
As time progresses, organic components tend to degrade and become stable, whereas conserva-
tive elements such as various heavy metals, chloride, sulfide and so on, will remain long after
waste stabilization occurs. Metals are found in high concentration in leachate as they are usually
precipitated within the landfill. The disposal of waste generated from domestic and industrial
areas makes landfill sites a necessary component of a metropolitan life cycle. However, low-lying
disposal sites which are lacking in proper leachate collection systems, observation of landfill gas
and collection tools, are a potential threat to underlying groundwater resources.
Landfill sites are complex environments characterized by many interacting physical, chem-
ical and biological processes. Leachate from landfills exhibits a major potential environmen-
tal impact for groundwater and surface water pollution and represents a potential health
risk to both surrounding ecosystems and human populations. Mangaluru generates around
250 tons of solid waste every day, of which 200 tones is collected and disposed in the landfill
located at Vamanjoor. Vamanjoor, which is located 15 km from the city, is along a national
185

Figure 1. Google Earth image of the study area.
highway (NH13) and is home to many educational institutes. The dumping yard has an area
of 28.32 hectares which is poorly managed. In the vicinity, the ground water is getting pol-
luted because of contamination by leachate.
The groundwater flow model is simulated with the help of Visual MODFLOW, and the
movement of the contaminant subjected to a variety of boundary conditions has been simu-
lated by using MODFLOW, MT3DMS and MODPATH which are widely used (Rejani et al.,
2008; Da An et al., 2013). All the complex processes such as advection, dispersion and chemi-
cal reaction are well addressed with the software.
2 STUDY AREA
Vamanjoor is located in the Gurupur basin which covers an area of 841 km2 (Figure 1). The
river Gurupur is a major river flowing in a westerly direction in the Dakshina Kannada district
in Karnataka State. The basin covers the foothills of the Western Ghats; in the middle portion
lies the lateritic plateaus and a flat coastal alluvium at its mouth. The area lies between 12°50’ to
13°10’ north latitude and 74°0’ to 75°5’ east longitude. The study area is tropical, with a humid
type of climate and gets an annual average precipitation of 3,810 mm. From the previous stud-
ies conducted in the area, the aquifer is categorized as unconfined with rich lateritic formation
and having good groundwater potential (Harshendra, 1991). The transmissivity of the aquifer
and its specific yield was determined to be 10 and 213 m2/day and 7.85%, respectively.
3 METHODOLOGY
3.1 Conceptual model development

To make the simulation model, a finite difference codes MODFLOW 2000 (Harbaugh et al.,
2000) a constant density flow model and MT3DMS 5.2 (Zheng, 2006; Zheng & Wang, 1999)
were used to solve the space and time-based heads of groundwater and concentration of
186

Figure 2. Location of the subbasin.
chloride of the subbasin. The model domain is enclosed by 1,622 active cells which includes
the total area with cell dimensions of 100 m × 100 m. The domain extends 30 m in the vertical
direction which is representing a mono layered unconfined aquifer. In order to make a dig-
ital elevation model (DEM), the contour lines of the toposheets numbered 48/L/13/NE and
48/L/13/NW (scale 1:25000) were considered. The spatial discretization details of the model are
given by X or longitudinal direction 485640 E and Y or latitudinal direction 1427956 N (with
respect to origin of UTM WGS 1984 Zone 43), the number of cells in X and Y direction being
41 (Figure 2). For modeling purposes, everyday time step was used. The discretization based
on time as well as space was arrived at after the first stages which were based on the precision
of the results.
3.2 Input parameters and boundary conditions

As there was some difficulty in obtaining data, whatever data were available were col-
lected and the model was built in the most appropriate way. The rivers were assigned with
an arc feature and assigned a river boundary condition using the RIV package which is
available in MODFLOW. The northern boundary was assigned with the river boundary
for which data such as river bed conductance, its thickness and stage data were assigned
as per available data from previous research and field visits. The Drain (DRN) package
of MODFLOW was used to simulate the effect of a drainage network in the model. All
boundaries, other than the northern boundary, were assigned as no flow boundaries or
impermeable boundaries. They were simulated by specifying cells for which a flow equa-
tion is not solved. No flow cells were used to delete the portion of the array of cells
beyond the aquifer boundary.
The annual average rainfall in the region is about 3,810 mm (Figure 3), which is from the
nearest rainfall station located at Bajpe airport at a radial distance of 5 km from the landfill.
Rainfall recharge is considered as the main source of replenishment for the aquifer. From
earlier studies, it was found that the recharge co-efficient appropriate to this region varies
from 10 to 20%. Hence, for the present study, a groundwater recharge co-efficient of 20% was
taken. A total number of 68 wells were considered in the model domain which was based on
available data. The water drawn per well was calculated based on the irrigation requirements
and also the domestic needs.
The chloride concentration was taken as the indicator of pollution due to leachate because
it is considered as the most consistent parameter in order to find leachate contamination
(Papadopoulou et al., 2007). The concentration of chloride measured from landfill leachate
during October 2016 was analyzed and the value assigned to each cell of the landfill and
taken as an initial concentration for the transport model. The concentration of chloride in
the groundwater samples collected from observation wells was analyzed.
187

Figure 3. Annual rainfall data obtained from Bajpe airport (Obtained from IMD, Bangalore).
3.3 Model simulation

The starting values assigned for the dependent variable, such as freshwater head for groundwa-
ter flow and concentration for the solute transport were represented by the initial condition.
In the present study, steady state calibration was carried out for October 2016 and simulated
water levels were assigned as initial conditions. First, the aquifer parameters were assigned in
a random manner and calibration was achieved by a trial and error method until a satisfac-
tory match was reached between the observed as well as the simulated head. The head of
groundwater head got as a result of steady state simulation is compared with the one third of
the data of observation well got from the subbasin and also the record of previous research of
Honnanagoudar (2015). The obtained head is taken as the initial head condition for transient
groundwater flow simulation done as a basis of daily time.
4 RESULTS
The scatter plot of observed and model simulated values were plotted with the x and y axes
having the same intervals and a 1:1 trend line (or 45° line) was fitted diagonally at point (0,0)
across the plot area (Figure 4). The figure reveals that the model fits the observed groundwater
heads as all points are lying close to the diagonal line. The RMSE value for the observed and
simulated values of head was obtained as 0.325 m, which also reveals that it is a perfect fit and
can be used for further applications. The groundwater head thus obtained was compared with
a one third value of observation wells of the subbasin and also with the data from previous
research (Honnanagoudar, 2015). The graph showed a convincingly good agreement between
the observed and simulated head with a RMSE value of 0.625 m. The process of validation
was then carried out by taking the water head obtained from the three observation wells in the
area maintained by the Central Ground Water Board and the Department of Mines and Geol-
ogy, Government of Karnataka. An RMSE value of 1.15 m was obtained after analyzing the
observed and calibrated groundwater head. The results were found to be consistent with that
of the calibrated results and therefore, the model was considered to be reliable for future pre-
diction. After successful calibration and validation, parameters such as recharge co-efficient
10%, porosity 30%, bed conductance of 15 m/day, horizontal conductivity 7 m/day, specific
yield 7.85% and transmissivity of 213 m2/day were taken for future application of the model.
The hydrodynamic dispersivity was given initially and adjusted by a trial and error method
during calibration. Horizontal transverse dispersivity of one tenth was suggested by Cobaner
et al. (2012). As per Bhosale and Kumar (2001) for related coastal aquifer conditions, the value
of longitudinal dispersivity can be taken between 15 to 150 m. Similar to the calibration of the
188

Figure 4. Scatter plot of steady state calibration.
Figure 5. Simulated chloride levels which

shows the spread of the contaminant Figure 6. Simulated chloride levels for
around the landfill. maximum recharge level.
flow parameters, calibration of the transport parameter was also completed. Only four wells
were located in the contaminated region which were chosen as observation wells. The chloride
level of the landfill leachate was identified as 6,000 mg/l during October 2016. The chloride
level of water in the observation wells was analyzed and compared with the simulated values.
After successful calibration, the transport parameter such as longitudinal and transverse dis-
persivity was taken as 25 m and 2.5 m, respectively. The model was validated by comparing the
values of that maintained by the Karnataka State Pollution Control Board during 2016. Since
the results were reliable, the model was taken for forecasting the future scenario.
4.1 Predictive simulation

An effort was made to determine the pollution due to landfill as shown in Figure 5, which
indicates that the pollutant has spread for a distance of 1 km around the landfill. An attempt
was also made to forecast a scenario for maximum rainfall. A maximum annual rainfall of
4.75 m had occurred in 2001 in the study area in the last 15 years (Figure 6). The resulting
simulation shows that the spread of the contaminant will reach 1.8 km around the landfill
during such heavy rainfall.
189

5 CONCLUSION
Groundwater, which is one of the world’s most important natural resources, is under constant
threat due to various human activities. The purpose of the present study is to understand the
extent of contamination due to landfill by using the software MODFLOW and MT3DMS.
After successfully calibrating the model with a RMSE value for observed and simulation
head equal to 0.32 m, the model was applied for predicting various scenarios. The results of
the study show that the spread of the contaminant has reached almost a 1 km radius around
the landfill. Additionally, the model predicted that contamination could reach a distance of
1.8 km when the value of maximum annual rainfall of 4.75 m is included. Since the area of
the current study is home to many educational institutes and also a residential area, urgent
attention must be given to prevent the spread of the contaminants.
REFERENCES
Bhosale, D.D. & Kumar, C.P. Simulation of seawater intrusion in Ernakulam coast. Retrieved from:
http://www.angelfire.com/nh/cpkumar/publication/ernac.pdf. Accessed: 10/10/2017.
Cobaner, M., Yurtal, R., Dogan, A. & Motz, L.H. (2012). Three dimensional simulation of seawater
intrusion in coastal aquifers: A case study in the Goksu Deltaic Plain. Journal of Hydrology, 262–280.
Da An, Yonghai Jiang, Beidou Xi, Zhifei Ma, Yu Yang Queping Yang Mingxiao L, Jinbao Zhang
Shunguo Ba & Lei Jiang (2013). Analysis for remedial alternatives of unregulated municipal solid
waste landfills leachate-contaminated groundwater front. Earth Sci. 7(3), 310–319.
Dinesh Kumar Babu J. Alappat (2005). Evaluating leachate contamination potential of landfill sites
using leachate pollution index. Clean Technologies and Environmental Policy, 7, 190–197.
Harbaugh A.W., Banta, E.R., Hill, M.C. & Mc Donald, M.G. (2000). MODFLOW – 2000, the US Geo-
logical Survey Modular Groundwater Model—User guide to modularization concepts and ground
water flow process. US Geological Survey, Open File Report, 1–92.
Harshendra K (1991). Studies on water quality and soil fertility in relation to crop yield in selected river basins
of D.K District of Karnataka State (Ph.D. Thesis, Mangalore University, Karnataka, India, 146–147).
Honnanagoudar, S.A. (2015). Studies on aquifer characterization and seawater intrusion vulnerability
assessment of coastal Dakshina Kannada district, Karnataka (Ph.D. Thesis, National Institute of
Technology Karnataka, Surathkal India, 184–185).
Papadopoulou, M.P., Karatzas, G.P. & Bougioukou, G.G. (2007). Numerical modeling of the environ-
mental impact of landfill leachate leakage on groundwater quality – a field application. Environment
Modeling and Assessment, 12, 43–54.
R. Rejani, Madan K. Jha S.N. Panda R. Mull (2008). Simulation modeling for efficient groundwater
management in Balasore Coastal Basin, India. Water Resources Management, 22(1), 23–50.
Zheng, C. (2006). MT3DMS v 5.2. Supplemental user’s guide. Department of Geological Sciences, Uni-
versity of Alabama.
Zheng, C. & Wang, K. (1999). A modular three dimensional multispecies transport model for simulation of
advection, dispersion and chemical reactions of contaminants in groundwater systems. Contract Report
SERD 99–1, U.S. Army Corps of Engineers, United States.
190

Critical review of water quality modeling and farm scale nutrient

transport models widely used
Nishchhal Nihal Pandey, Vishnu Sharma & H.N. Udayashankar

Manipal Institute of Technology, Manipal, India
ABSTRACT: Use of software engineering concepts in environmental modeling initiated

many researchers to develop new models to predict the future scenario in water quality man-
agement. To understand the basics in numerical water quality modeling, a general differential
equation involving advection, diffusion and source term is explained along with a modeling
framework. We reviewed the present framework of farm-scale nutrient transport models that
are used to simulate and predict the nutrient concentration in each water body. Each model
addresses the loss of nitrogen and phosphorus from rural population and nutrient load and
concentration in waterways. Some of the key farm scale nutrient transport models that are
widely used at the present time are discussed.
Keywords: Water quality modeling, Nutrient load, Nutrient transport
1 INTRODUCTION
At present, the environmental management has become more complex (Purandara et al.
2012). This is mainly due to anthropogenic activities, rapid urbanization and fleeting growth
in agricultural fields, environmental mandates, recreational interests, hydropower generation,
over-allocation and land use patterns which results in climate change and fragmented nature
of available information (Purandara et al. 2012) (Welsh et al. 2013). Water quality modeling
of freshwaters is a trending research area in the present scenario which focuses on the evalu-
ation of biological and chemical status of the water bodies (Altenburger et al. 2015). Point
source pollution is found to be managed well compared to non-point source pollution. Non-
point source pollutants do not originate from a statutory point source, but dispersed into the
receiving water by various means. NPS pollutants include components from evapotranspi-
ration, percolation, interception, absorption, and vegetative covers which can bring about
changes in the hydrologic cycle, water balance, land surface characteristics and surface water
characteristics (Tong & Chen 2002) (LeBlanc et al. 1997) (Lai et al. 2011).
Water quality models are effective tools for understanding the fate and transport of contami-
nants in a river system (Wang et al. 2013). Many site-specific river basin models were developed
and used by the engineers in the early days for decision making purpose. Numerous general
water quality models were developed in the recent years for understanding various hydrological
processes. Some of them include, QUAL 2 K (Esterby 1996), WASP7, CE-QUAL-ICM (Chuco
2004) (Bahadur et al. 2013) (D., McKinnon, A., Brinkman, R., Trott, L., Undu, M., Muawanah
and Rachmansyah (2009)) (Cameira, M., Fernando, R., Ahuja, L. and Ma, L. (2007)), HEC-
RAS, MIKE11, DUFLOW, AQUASIM, DESERT, EFDC model (U.S. Environmental Protec-
tion Agency 1999) (The U.S. Environmental Protection Agency 1997), GSTAR-1D, CASC2D.
Computer models used in integrated water quality modeling are capable of combining
various spatial and environmental data for complex studies. To understand these water qual-
ity models, a basic knowledge on mathematical modeling is required. For example, mod-
els have been developed to know the fate of organic or inorganic contaminants transport
or to know the transport of nutrients, pesticide, sediment loss, erosion for informing land
191

management practices including supplying rates and the application of fertilizer and irriga-
tion; and for both urban and rural outlines. In the present study, the scope is limited to the
models currently used for modeling nitrogen and phosphorus (these are the parameters of
concern in most of the circumstances) loss from rural land and the concentrations and loads
in freshwater.
Process-based models are extensively used to inspect nutrient dynamics for water manage-
ment purposes. Simulating nutrient transport and transformation processes from agricultural
land into water bodies at the catchment scale are particularly relevant and challenging tasks
for water authorities (Tuo et al. 2015) (Hashemi, F., Olesen, J., Dalgaard, T. and Børgesen,
C. (2016)).
2.1 Basic model development

All the models are primarily based on the principle of conservation of mass. For a variable
within a control volume, a one-dimensional advection-dispersion equation with source term
can be derived.
∂C ∂C ∂ 2C
= −U + E 2 + Sc (1)
∂t ∂x ∂x
where, t is time, U is the steady-state average velocity of the water in the flow direction (x)
and E is the dispersion coefficient. U and E are assumed to be in the direction of flow (x) of
the water body. In a great number of water bodies, eq. (1) can be applied where flow can be
approximated as one-dimensional. First term represents the rate of change of concentration
(C) of a component, second term represents advection (first order), third term represents
dispersion (second order) and the fourth term represents the source term which incorpo-
rate inflows, outflows, and reactions due to physical, chemical, and/or biological processes
(Stamou & Rutschmann 2011).
Eq. (1) can be solved numerically using finite difference method. The source term may
vary according to the type of water body under consideration (Stamou & Rutschmann 2011)
(Gelda & Effler 2000).
2.2 Assumptions and inputs for the model

All mathematical models for environment studies are generated based on certain practical sys-
tems or equations which means that it can only accept a particular set of data as inputs. More
over most of the mathematical models have to be developed through computers, a valid coding
program is needed to be generated to process the input for the model. Before developing any
mathematical framework there are a set of assumptions which are made and the mathematical
model is valid only when these sets of assumptions are strictly followed. The assumptions made
can be of Numerical in nature which are made during the implementation or problem solving
part of the mathematical model or logical/qualitative type which is made during the generation
of the equation or the model itself. As we make all our calculations based on the assumptions
made, it is of utmost importance that the assumptions are followed during the calculations part
or the model becomes insignificant. (Shukla, j., hallam, t. g. and capasso, v) (holzbecher, e. o)
(wang, q., li, s., jia, p., qi, c. and ding, f) (A. Mudgal, C. Baffaut, S. H. Anderson, E. J. Sadler
and A. L. Thompson (2010)), (Cherry, K., Shepherd, M., Withers, P. and Mooney, S. (2008)).
2.3 Farm scale nutrient models

Farm scale models consider nutrient loss in reaction to land management decisions on a
farm, paddock, or plot scale. Sometimes called “root zone” models, these models calculate
192

Figure 1. Principal components of modeling framework.
the quantity of nutrients that are lost from the top-most soil layers instead of being absorbed
by plant roots.
2.3.1 FARMSIM
The model is developed especially for the complex, dynamic systems with many interacting
biophysical components along a wide range of soil, climatic and socioeconomic conditions
(Wijk et al. 2004).
2.3.2 NUANCES-FARMSIM
Nutrient Use in ANimal and Cropping systems: Efficiencies and Scales-FARM SIMula-
tor is an integrated crop—livestock model developed to analyse African smallholder farm
systems.
2.3.3 SWATRE
A transient one-dimensional finite-difference model for vertical unsaturated flow with water
uptake by roots is presented. In the model a number of boundary conditions are given for
both top and bottom of the system. At the top, 24-hr. data on rainfall, potential soil evapora-
tion and potential transpiration are needed. When the soil system remains unsaturated, one
of three bottom boundary conditions can be used: pressure head, zero flux or free drainage.
When the lower part of the system remains saturated, one can either give the groundwater
level or the flux through the bottom of the system as input. In the latter case the groundwater
level is computed (Belmans et al, 1983).
2.3.4 ANIMO
Simulation of the nitrogen behaviour in the soil and the nitrogen uptake by winter wheat was
performed using the model ANIMO (Rijtema & Kroes, 1991). It is a detailed process ori-
ented simulation model for evaluation of nitrate leaching to groundwater, N- and P-loads on
surface waters and Green House Gas emission. The model is primarily used for the ex-ante
evaluation of fertilization policy and legislation at regional and national scale. The output of
SWATRE model is taken as an input for the ANIMO mode.
2.3.5 OVERSEER
The OVERSEER nutrient budget model is a decision support tool (Wheeler et al. 2003) to assist
farmers and consultants develop nutrient plans (Wheeler et al. 2006). It produces estimates of
long-term average nutrient losses via drainage and runoff at a farm and farm block level. Over-
seer also estimates greenhouse gas emissions and aids in planning fertiliser applications.
193

Figure 2. Classification of farm scale models based on intended use and the question of focus.
2.3.6 AMAIZEN
The AmaizeN is a maize growth simulation software. The core of the system is the daytime
step simulation model of maize growth and development, driven by solar radiation and inter-
acting with soils. This model is an extension of the maize growth simulation model which is
modified for cooler situations.
2.3.7 APSIM
Agricultural Production System Simulator is a software system which allows models of crop
and pasture production, residue decomposition, soil water and nutrient flow, and erosion to
be readily re-configured to simulate various production systems and soil and crop manage-
ment to be dynamically simulated using conditional rules (Mccown et al. 1996). The main
objective behind the development of the model was to simulate biophysical process in farm-
ing systems, in particular where there is interest in the economic and ecological outcomes of
management practice in the face of climatic risk (Keating et al. 2003). The model simulates
the dynamics of soil-/plant-management interactions within a single crop or a cropping sys-
tem (Wang et al. 2002).
2.3.8 MitAgator
MitAgator is a farm scale GIS (Geographic Information System) based DST (Decision Sup-
port Tool) that has been developed to identify and estimate nitrogen, phosphorus, sediment,
and E. coli loss spatially across a farm landscape. This model is a spatially explicit model that
extends the results produced by Overseer to identify where on the farm property nutrient loss
is occurring (Anastasiadis et al. 2013).
2.3.9 SPASMO
Soil Plant Atmosphere System Model is a physics based model of plant growth and nitro-
gen leaching, with some estimation of phosphorus. It focuses on arable and 15 horticultural
activities. SPASMO produces estimates of crop production, drainage, and nutrient leaching
from soil on a daily time step. Its inputs include spatial data on climate, irrigation, soil prop-
erties, land use, and crop and stock management (Anastasiadis et al. 2013).
3 CONCLUSIONS
A brief description of numerical water quality models has been explained. Farm scale models
and the use of models for understanding nutrient loss to waterways have been discussed. The
practice of modelling is well established among scientists as it provides an effective way to
think about and understand complex phenomena. Models provide structure to guide new
194

research by combining existing knowledge and identifying areas where new knowledge is
needed. In this way, models are a summary of the science and a guide for further science.
Models are established, used and sophisticated in the context of a wider scientific com-
munity. This community both updates the design of models and tests the validity of their
results. This is an integral part of the accepted processes by which scientific understanding
improves.
REFERENCES
Argent, R., Sojda, R., Giupponi, C., McIntosh, B., Voinov, A., & Maier, H. (2016). Best practices for
conceptual modelling in environmental planning and management. Environmental Modelling & Soft-
ware, 80, 113–121. http://dx.doi.org/10.1016/j.envsoft.2016.02.023.
Bahadur, R., Ziemniak, C., Amstutz, D., & Samuels, W. (2013). Global River Basin Modeling and
Contaminant Transport. American Journal of Climate Change, 02(02), 138–146. http://dx.doi.
org/10.4236/ajcc.2013.22014.
Belmans, C., Wesseling, J., & Feddes, R. (1983). Simulation model of the water balance of a cropped soil:
SWATRE. Journal of Hydrology, 63(3–4), 271–286. http://dx.doi.org/10.1016/0022-1694(83)90045-8.
Cameira, M., Fernando, R., Ahuja, L. and Ma, L. (2007). Using RZWQM to simulate the fate of nitro-
gen in field soil–crop environment in the Mediterranean region. Agricultural Water Management,
90(1–2), pp.121–136.
Cherry, K., Shepherd, M., Withers, P. and Mooney, S. (2008). Assessing the effectiveness of actions
to mitigate nutrient loss from agriculture: A review of methods. Science of the Total Environment,
406(1–2), pp.1–23.
Cunderlik, J. (2007). River Bank Erosion Assessment using 3D Hydrodynamic and Sediment Transport
Modeling. Journal of Water Management Modeling. http://dx.doi.org/10.14796/jwmm.r227-02.
Esterby, S. (1996). Review of methods for the detection and estimation of trends with emphasis on
water quality applications. Hydrological Processes, 10(2), 127–149. http://dx.doi.org/10.1002/
(sici)1099-1085(199602)10:2<127::aid-hyp354>3.0.co;2–8.
Gelda, R., & Effler, S. (2003). Application of a Probabilistic Ammonia Model: Identification of Impor-
tant Model Inputs and Critique of a TMDL Analysis for an Urban Lake. Lake and Reservoir Man-
agement, 19(3), 187–199. http://dx.doi.org/10.1080/07438140309354084.
Guagnano, D., Rusconi, E., & Umiltà, C. (2013). Joint (Mis-)Representations: A Reply to Welsh et al.
(2013). Journal of Motor Behavior, 45(1), 7–8. http://dx.doi.org/10.1080/00222895.2012.752688.
Gyawali, S., Techato, K., Monprapussorn, S., & Yuangyai, C. (2013). Integrating Land Use and Water
Quality for Environmental based Land Use Planning for U-tapao River Basin, Thailand. Procedia—
Social and Behavioral Sciences, 91, 556–563. http://dx.doi.org/10.1016/j.sbspro.2013.08.454.
Hashemi, F., Olesen, J., Dalgaard, T. and Børgesen, C. (2016). Review of scenario analyses to reduce
agricultural nitrogen and phosphorus loading to the aquatic environment. Science of the Total Envi-
ronment, 573, pp. 608–626.
Holzbecher, E. (2012). Environmental modeling. Heidelberg: Springer.
Keating, B., Carberry, P., Hammer, G., Probert, M., Robertson, M., & Holzworth, D. et al. (2003). An
overview of APSIM, a model designed for farming systems simulation. European Journal of Agro
nomy, 18(3–4), 267–288. http://dx.doi.org/10.1016/s1161-0301(02)00108-9.
McKinnon, D., A. Brinkman, R. Trott, L. Undu, M. Muawanah and Rachmansyah (2009). The fate
of organic matter derived from small-scale fish cage aquaculture in coastal waters of Sulawesi and
Sumatra, Indonesia. Aquaculture, 295(1–2), pp. 60–75.
Mudgal, A., C. Baffaut, S.H. Anderson, E.J. Sadler and A.L. Thompson (2010). APEX Model Assess-
ment of Variable Landscapes on Runoff and Dissolved Herbicides. Transactions of the ASABE,
53(4), pp.1047–1058.
Purandara, B., Varadarajan, N., Venkatesh, B., & Choubey, V. (2011). Surface water quality evalua-
tion and modeling of Ghataprabha River, Karnataka, India. Environmental Monitoring and Assess-
ment, 184(3), 1371–1378. http://dx.doi.org/10.1007/s10661-011-2047-1.
Rabi, A., Hadzima-Nyarko, M., & Šperac, M. (2015). Modelling river temperature from air temper-
ature: case of the River Drava (Croatia). Hydrological Sciences Journal, 60(9), 1490–1507. http://
dx.doi.org/10.1080/02626667.2014.914215.
Rijtema, P., & Kroes, J. (1991). Some results of nitrogen simulations with the model ANIMO. Fertilizer
Research, 27(2–3), 189–198. http://dx.doi.org/10.1007/bf01051127.
Shukla, J., Hallam, T., & Capasso, V. Mathematical modelling of environmental and ecological systems.
195

Stamou, A., & Rutschmann, P. (2011). Teaching simple water quality models. Education for Chemical
Engineers, 6(4), e132–e141. http://dx.doi.org/10.1016/j.ece.2011.08.005.
Wang, Q., Li, S., Jia, P., Qi, C., & Ding, F. (2013). A Review of Surface Water Quality Models. The
Scientific World Journal, 2013, 1–7. http://dx.doi.org/10.1155/2013/231768.
WANG Tian-Xiao (2013) Chemical constituents from Psoraleacorylifolia and their antioxidant
α-glucosidase inhibitory and antimicrobial activities. (2013). China Journal of Chinese Materia Med-
ica. http://dx.doi.org/10.4268/cjcmm20131423.
196

A novel hybrid material for the trace removal of hexavalent

chromium [Cr(VI)] from contaminated water
A.R. Laiju
National Institute of Technology, Srinagar, Uttarakhand, India
S. Sarkar
Indian Institute of Technology, Roorkee, Uttarakhand, India
ABSTRACT: This study investigated the performance of two anion exchange resins
and also synthesized and evaluated the performance of a hybrid material consisting of a
strong base anion exchange resin dispersed with FeS (iron sulphide) nanoparticles for the
removal of hexavalent chromium, Cr(VI). The synthesis of the hybrid was carried out by
using an in situ process, where the strong base anion exchange resin serves as a nanoreac-
tor and provides a confined medium for synthesis. They stabilize and isolate the synthe-
sized nanoparticles preventing their aggregation. Equilibrium batch studies, adsorption
isotherm and column studies were performed to determine the maximum uptake capacity
for Cr(VI). Comparison of a fixed bed column run between the hybrid material and par-
ent resin confirmed that the Cr(VI) was selectively removed and the hybrid showed higher
capacity. The wide availability of resin and low-cost chemicals for synthesis and regen-
eration will make hybrid material an attractive option for the removal of Cr(VI) from
contaminated water.
1 INTRODUCTION
Chromium is unique among toxic heavy metals in the environment and toxicity is regu-
lated on the basis of its oxidation state and total concentration. In water distribution
systems and in water treatment processes, chromium exists mainly as trivalent chromium,
Cr(III) and hexavalent chromium, Cr(VI) (Kimbrough et al., 1999). Cr(VI), an inorganic
contaminant which received public attention recently, can be considered as a potential
human carcinogen (International Agency for Research on Cancer, 2012). Public concern
and potential adverse health effects prompt the investigation of Cr(VI) in drinking water
supplies below Maximum Contaminant Level (MCL). In India, the maximum permis-
sible limit for Cr(VI) in drinking water is 50 µg/L. Most of the regulations by different
agencies are based on total chromium not just Cr(VI), due to the absence of sufficient
risk analysis for the oral intake of Cr(VI) (Anderson, 1997). To reduce the contamination
of Cr(VI) from water and wastewater, various methods have been used mainly involv-
ing reduction followed by precipitation, adsorption, ion exchange and membrane process
(Sharma et al., 2008).
The objectives of the present study are to evaluate the performance of a strong base anion
exchange resin, Amberlite IRA 400 (IRA 400), a weak base anion exchange resin, Lewatit
MP 64 (LMP 64) and to synthesize a novel Hybrid Ion exchange Material (HIM) and to
validate the performance of trace concentration of Cr(VI) from contaminated water. The
physical properties of the resin are shown in Table 1.
197

Table 1. Physical properties of resins.
Resin Amberlite IRA-400 Lewatit MP 64
Type Strong base anion exchange Weak base anion exchange

Functional group Quaternary ammonium Tertiary amine
Matrix Styrene–divinyl benzene (gel) Styrene–divinyl benzene
(macro porous)
Ionic form Chloride Chloride
Temperature limit (°C) 77 70
Moisture holding capacity (%) 40–47 50–60
Particle size (μm) 600–750 300–1250
Total exchange Capacity (eq//L) 1.4 1.3
2 EXPERIMENTAL METHODOLOGY
2.1 Synthesis of HIM

For the synthesis of HIM, iron sulphide (FeS) nanoparticles were immobilized in IRA
400 using an in situ synthesis process (Sarkar et al., 2012) via two steps as shown in Equa-
tions 1 and 2. The over bar denotes theres in phase. In total, 10 gm of IRA 400 was added
to 100 ml of 0.3 M Na2S solution and shaken in an orbital shaking incubator for one hour
at a temperature of 25 ± 0.5ºC and the resultant solution was filtered and washed thor-
oughly with distilled water. The resin was then transferred into a solution containing 30 g/L
of FeCl2 solution of 100 ml and shaken for one hour at a temperature of 25 ± 0.5ºC. The
resultant material, now expected to be impregnated with FeS nanoparticles, was drained and
washed with distilled water and stored in an ethanol solution to prevent further oxidation.
Figure 1 shows a diagram of HIM containing FeS nanoparticles.
2R (CH3 )3 C1 + S 2 − ↔ R (CH3 )3  S 2 − + 2Cl − (1)

2
 R ( CH3 ) 3  S 2− + Fe 2+ + 2C1− ↔ 2R( CH3 ) 3 C1− + FeS ( s ) (2)

2
2.2 Batch experiments

In order to evaluate the performance of resin and hybrid material for Cr(VI) removal, the
effect of pH, initial concentration, resin dosage and competing anions (SO42−, Cl−, HCO3−)
were investigated. The removal efficiency, η (%) and the amount of Cr(VI) ions adsorbed
at equilibrium qe, (mg of Cr(VI)/gm of resin) can be calculated using Equations 3 and 4,
respectively, where Co and Ce are initial and equilibrium concentrations of Cr(VI), m is mass
of material added, V is the volume of solution.
C0 − Ce
η (%) = × 100 (3)
C0
V (C0 − Ce )
qe = (4)
m
2.3 Fixed bed column run

Fixed bed column studies were performed in a plexiglass column of 11 mm internal diameter
and 300 mm height. With the help of a microprocessor based peristaltic pump (Ravel RH
P100 L), feed water was passed through the column with IRA 400, LMP 64 and HIM at a
certain pH until breakthrough. Figure 2 shows a schematic diagram of a fixed bed column
run. Effluent samples were collected at definite intervals during the column run and were
198

Figure 1. Diagram of HIM containing FeS nanoparticles.
Figure 2. Schematic diagram for fixed bed column run.
analyzed for Cr(VI) concentration by the 1,5 diphenylcarbazide method by using a split beam
UV visible spectrophotometer (T60 UV, PG Instruments, UK) at 540 nm.
3.1 Characteristics of HIM

The characteristics of IRA 400 and HIM were investigated by SEM (Scanning Electron
Microscopy) with EDX (Energy-Dispersive X-ray analysis) as presented in Figure 3.
SEM and EDX results of HIM shows that the FeS particles were uniformly precipitated in
side the HIM and the Fe content was approximately 6% by weight, as revealed by the EDX study.
3.2 Effect of resin dosage

Increasing the dosage of resin and HIM, increased the efficiency of Cr(VI) removal. After
0.8 g/L, the increment in Cr(VI) removal was not increased as significantly. The increase in
removal efficiency is due to an increase in number of ion exchange sites for exchange between
the liquid phase and the exchanger phase (Babu & Gupta, 2008). If the dosage is increased
by keeping the initial concentration constant, the amount of Cr(VI) absorbed per unit mass
showed a decrease due to the unavailability of Cr(VI) ions for the exchange sites. The uptake
capacity dropped from 92.6 to 4, 45.27 to 2.38 and 107.5 to 5.32 mg/gm for IRA 400, LMP
64 and HIM, respectively, by increasing the dosage from 0.2 to 1.2 g/L (Figure 4). The reason
for the drop in the uptake capacity while increasing the dosage is due to ion exchange sites
remaining unsaturated during the exchange process.
3.3 Effect of pH
From Figure 5, the maximum uptakes of 11.08, 7.01 and 14.41 mg/gm were obtained for
IRA 400, LMP 64 and HIM at pH 4. For resin and hybrid, the uptake, as well as efficiency,
increases as the pH decreases.
199

Figure 3. SEM image of IRA 400 (A), HIM (B) and EDX of HIM (C).
Figure 4. Effect of resin dosage.
From the predominance diagram, H2CrO4 is predominant for a pH less than 1, HCrO4− for
a pH between 1 and 6.5 and divalent CrO42− for a pH above 6.5. Dimerization of HCrO4− ions
(Cr2O72−) is possible if the concentration is higher than 1 g/L (Saleh et al., 1989). The increased
Cr(VI) removal efficiency at an acidic pH is mainly due to the fact that HCrO4−, being mono-
valent, can attach to a single ion exchange functional group and CrO42−, a divalent, needs to
bind to two ion exchange functional groups (Sengupta & Clifford, 1986). At an alkaline pH,
the sorption trend is likely decreased due to the competition between CrO42− and OH− for the
binding sites on the exchanger which results in lower uptake. The type of functional group,
quaternary ammonium moiety of IRA 400, has a significant effect on the uptake of Cr(VI)
rather than the tertiary amine of LMP 64 (Pehlivan & Cetin, 2009; McGuire et al., 2007).
3.4 Effect of initial concentration

Initial Cr(VI) concentration varied from 1 mg/L to 15 mg/L at pH 7 with a dosage of 0.8 g/L
and the results are shown in Figure 6. Maximum uptakes of 4.71, 2.8 and 6.07 mg/gm were
obtained for IRA 400, LMP 64 and HIM, respectively, for an initial concentration of 1 mg/L
and as the initial concentration of Cr(VI) increases, the Cr(VI) uptake decreases gradually. At
lower concentrations, all Cr(VI) molecules present in the solution interact with the binding
sites of the resin and as the concentration increases, the binding sites become saturated lead-
ing to decreased Cr(VI) removal efficiency (Balan et al., 2013; Neagu et al., 2003).
3.5 Effect of competing anions

The main competing anions for Cr(VI) in ground water, SO42−, Cl− and HCO3, were investi-
gated under different concentrations from 0 to 200 mg/L. Chloride and bicarbonate did not
200

interfere in Cr(VI) removal, so the result is not included but as the concentration of sulfate
increased the uptake capacity decreased, showing that the affinity for sulfate increases as the
concentration of sulfate increases for IRA 400, LMP 64 and HIM (Figure 7).
3.6 Adsorption isotherm

Adsorption capacities can be described by using Langmuir and Freundlich models indicated
by Equations 5 and 6, and the results are shown in Figure 8.
1 1 1
= + (5)
q e Q max bC e Q max
1
lo qq e = log C e + lo qk f (6)
n
Figure 5. Effect of pH.
Figure 6. Effect of initial concentration.
Figure 7. Effect of sulfate.
201

where Qmax is the maximum uptake capacity of Cr(VI) (mg/gm), b is the Langmuir constant
(g/L), kf is the Freundlich constant (mg/g), and 1/n is the adsorption intensity. The correla-
tion coefficients (R2 value) of both Langmuir and Freundlich models are close to 1, but the
Freundlich model shows better correlation for the whole initial high concentration of Cr(VI)
than the Langmuir model. Thus, it indicates that the adsorption is not a monolayer layer
adsorption, rather it is a multilayer adsorption. The parameters obtained for the Langmuir
and Freundlich models are shown in Table 2.
3.7 Kinetic study

From Figure 9, observed that during the initial stage the rate removal is faster and further
it took longer time to reach equilibrium after 360 minutes from the start of experiment.
Interdiffusion of ions take place in two steps, (1) liquid film diffusion and (2) intraparticle
Figure 8. (a) Langmuir isotherm (b) Freundlich isotherm.
Table 2. Parameters for Langmuir and Freundlich isotherm.
Langmuir isotherm Freundlich isotherm
Qmax (mg/g) b R2 kf (mg/g) n R2
IRA 400 8.143 29.24 0.8742 7.72 2.68 0.9869

LMP 64 7.183 0.4 0.8384 1.42 1.50 0.956
HIM 10.93 130.7 0.9466 27.97 2.19 0.978
Figure 9. Fractional uptake of aqueous Cr(VI).
202

diffusion and slowest among the two process will control the overall exchange process.
(Li & Sengupta, 2000; DeMarco et al., 2003; Greenleaf et al., 2006; Chanthapon et al., 2017).
Normalized fractional loading or uptake (F) by the exchanger during the batch kinetic
study changes with time (Figure 9) and can be defined by Equation 7, where Ccr,o, Ccr,t, and
Ccr,e represent concentration of Cr(VI) at initial, timet and equilibrium respectively and qcr,t,
qcr,e are Cr(VI) uptake at time t and at equilibrium respectively.
F=
qCr ,t
=
(CCr , 0 − CCr ,r ) (7)
qcr .e (CCr.0 − CCr.e )
3.8 Fixed bed column run and regeneration

Figure 10 shows the effluent concentration of Cr(VI) during separate column runs with IRA
400, LMP 64 and HIM, with identical influent characteristics. Effluent Cr (VI) concentration
of 50 µg/L occurred after approximately 2100, 650 and 5500 bed volumes (BVs, volume of
water treated per volume of resin) and exhausted at 4275, 1532 and 10268 BVs for IRA 400,
LMP 64 and HIM respectively. Interruption test was carried out during the fixed column run
for HIM, marked with circles in Figure 9. When the flow resumed, the effluent concentra-
tion of Cr(VI) dropped significantly to 48.2 µg/L and it took approximately 1,250 BVs for
the effluent concentration to reach the concentration present prior to the interruption. The
interruption allows the sorbed C(VI) to evenly spread out inside the host ion exchanger.
Faster uptakes occurred after the interruption, providing evidence in support of intraparticle
diffusion being the rate limiting step for the hybrid material.
The mechanism behind the superior removal of Cr(VI) by HIM may be due to the high
positive charge of the functional group of the parent anion exchanger, Cr(VI) species either
HCrO4− or CrO42−, will be attracted inside the host material due to the Donnan membrane
effect. The Cr (VI) species inside the HIM undergo redox reaction with the impregnated iron
sulphide nanoparticles and this reaction frees up the active ion exchange site were Cr (VI)
initially occupied and can sorb new set of Cr (VI). Equations 8 and 9 show the redox reac-
tions between FeS nanoparticles and Cr (VI) inside the host material.
3FeS + 8HCrO4− + 32 H + → 3Fe3 + + 8Cr 3 + + 3SO42 − + 20 H 2O (8)

Cr 3 + + 3O H − → Cr (OH )3 (9)
Figure 10. Fixed bed column run.
203

Regeneration of exhausted resin and HIM were carried out by using 5% NaCl. Results
show that regenerated hybrid material (Re-HIM) was able to remove 5,869 BVs up to 175
µg/L effluent Cr(VI) concentration.
4 CONCLUSION
Iron sulphide impregnated hybrid material were synthesized by in-situ synthesis process on func-
tional polymer supported host strong base anion exchangeres in. The removal process was found
to be strongly dependent on the pH, dosage, initial concentration and sulfate concentration.
HIM shows high selectivity toward Cr(VI), with a high adsorption capacity found by fitting the
Freundlich model. A fixed bed column containing HIM with feed water composition of 200
µg/L of Cr(VI), sulphate, chloride and bicarbonate of 100 mg/L can treat up to 5,500 BVs before
reaching Cr (VI) concentration corresponding to Indian standards (50 µg/L). The column study
identified that the rate limiting step is the intraparticle solute transport mechanism. A fixed bed
column study and the possibility of regeneration shows that HIM can be considered as a prom-
ising hybrid material for the trace removal Cr(VI) from contaminated water.
REFERENCES
Anderson, R.A. (1997). Chromium as an essential nutrient for humans. Regulatory Toxicology and
Pharmacology, 26(26), s35–s41.
Babu, B.V. & Gupta, S. (2008). Adsorption of Cr(VI) using activated neem leaves: Kinetic studies.
Adsorption,14(1), 85–92.
Balan, C., Volf, I. & Bilba, D. (2013). Chromium (VI) removal from aqueous solutions by purolite
base anion-exchange resins with gel structure. Chemical Industry and Chemical Engineering Quar-
terly,19(4), 615–628.
Chanthapon, N., Sarkar, S., Kidkhunthod, P. & Padungthon, S. (2017). Lead removal by a reusable gel cat-
ion exchange resin containing nano-scale zero valent iron. Chemical Engineering Journal, 331, 545–555.
DeMarco, M.J., Sengupta, A.K. & Greenleaf, J.E. (2003). Arsenic removal using a polymeric/inorganic
hybrid sorbent. Water Research, 37(1), 164–176.
Greenleaf, J.E., Lin, J.C. & Sengupta, A.K. (2006). Two novel applications of ion exchange fibers:
Arsenic removal and chemical-free softening of hard water. Environmental Progress, 25(4), 300–311.
International Agency for Research on Cancer. (2012). IARC Monograph: Chromium (VI) Compounds.
147–168.
Kimbrough, D.E., Cohen, Y., Winer, A.M., Creelman, L. & Mabuni, C. (1999). A critical assessment of
chromium in the environment. Critical Reviews in Environmental Science and Technology, 29(1), 1–46.
Li, P. & Sengupta, A.K. (2000). Intraparticle diffusion during selective ion exchange with a macropo-
rous exchanger. Reactive and Functional Polymers, 44(3), 273–287.
McGuire, M.J., Blute, N.K., Qin, G., Kavounas, P., Froelich, D. & Fong, L. (2007). Hexavalent chro-
mium removal using anion exchange and reduction with coagulation and filtration. Water Research
Foundation, Project #3167, 140.
Neagu, V., Untea, I., Tudorache, E. & Luca, C. (2003). Retention of chromate ion by conventional and
N-ethylpyridinium strongly basic anion exchange resins. Reactive and Functional Polymers, 57(2–3),
119–124.
Pehlivan, E. & Cetin, S. 2009. Sorption of Cr(VI) ions on two Lewatit-anion exchange resins and their
quantitative determination using UV-visible spectrophotometer. Journal of Hazardous Materials,
163(1), 448–453.
Saleh, F.Y., Parkerton, T.F., Lewis, R.V., Huang, J.H. & Dickson, K.L. (1989). Kinetics of chromium
transformation in the environment. Science of the Total Environment, 86, 25–41.
Sarkar, S., Guibal, E., Quignard, F. & Sengupta, A.K. (2012). Polymer-supported metals and metal
oxide nanoparticles: Synthesis, characterization, and applications. Journal of Nanoparticle Research,
14(2), 1–24.
Sengupta, A.K. & Clifford, D. (1986). Important process variables in chromate ion exchange. Environ-
mental Science & Technology, 20(2), 149–55.
Sharma, S.K., Petrusevski, B. & Amy, G. (2008). Chromium removal from water: A review. Journal of
Water Supply: Research and Technology—AQUA, 57(8), 541–553.
204

Solid waste management practices and decision-making in India
R. Rajesh & B.K. Bindhu

Rajiv Gandhi Institute of Technology, Kottayam, Kerala, India
ABSTRACT: One of the most significant problems in urban India is poor Solid Waste
Management (SWM) systems that are leading to environmental, health and economical
issues. The aim of this paper is to provide an insight into the SWMs and practices in India,
and the challenges faced. The review focuses on three major areas: solid waste characteris-
tics; SWM systems; and decision-making. The two major issues at source are the increasing
amount of waste, and the changing composition of the waste streams. The issues in waste
logistics are due to poor waste segregation, unregulated waste transportation, and incom-
patible and outdated transportation technologies. Sustainable waste processing and disposal
practices are yet to be adopted and are facing numerous implementation and operational
challenges. There is a need to understand the operational issues in waste processing and dis-
posal so that better technological development and adoption at a macro level is made possi-
ble. The decision-making practices and solutions generated in India are unscientific or based
on piece meal approaches. Inefficiencies in waste management systems are primarily due to
poor alignment between requirements, constraints and operations in the SWM system. A sci-
entific SWM design that meets the fundamental requirements to handle a variety of waste, to
be environmentally friendly, and is economically and socially acceptable, is needed. A robust
SWM system design places emphasis on active public participation in decision-making, and
operational integration of its subcomponents.
1 INTRODUCTION
Solid waste generation, and its management, is a worldwide problem. In India, it is antici-
pated that about 260–300 million tons of solid waste per annum will be generated by 2050
(Joshi & Ahmed, 2016). This is primarily due to population growth, increasing urbanization
and socio-economic development. Solid waste leads to different environmental problems and
ecosystem changes, human health issues, and socio-economic issues. There is an increasing
focus on using Solid Waste Management (SWM) systems in the movement toward an envi-
ronmentally sustainable society. Over the last decades, there have been continuous changes
in the solid waste management practices evolving from the simple form of collection and
dumping, to integrated solid waste management arising from learned experiences. However,
there is a need to understand the operational issues too.
The aim of this paper is to provide an insight into the SWM systems and practices in
India.
2 LITERATURE REVIEW
Peer-reviewed articles published since 2000 from Scopus and Google Scholar databases
were selected using the keywords ‘e-waste’, ‘municipal solid waste’, ‘solid waste manage-
ment’, ‘waste recycling’, ‘life cycle assessment’, ‘waste disposal’, ‘environment assessment’,
and ‘multi-criteria decision-making’. The articles were included after careful review of the
abstracts.
205

2.1 Solid waste characteristics
Solid waste is the waste generated from domestic, commercial, and construction activities.
The source of any waste is from the process or activity undertaken by an individual as in
their home or from an enterprise (e.g. hospital, hotels, and industries). Solid waste can be
classified based on origin, source, risk, characteristics or treatability. Solid waste consists of
organic matter, fine earth, paper, plastic, glass, metals, rubber, textiles, leather, chemicals and
inert materials. Types of waste include biodegradable waste, recyclable material, inert waste
matter, domestic, and hazardous waste (Joshi & Ahmed, 2016; Nandy et al., 2015; Sharholy
et al., 2008).
During the last decade, solid waste generation has doubled (CPCB, 2013). Per capita waste
generation in cities varies from 0.2 kg to 0.6 kg per day depending upon the size of popula-
tion, and is estimated to increase by 1.33% annually (Joshi & Ahmed, 2016). The Indian
industrial sector generates an estimated 100 million tons/year of non-hazardous solid waste
(EBTC, 2014). Annually, about 12 million tons of inert waste is generated in India from street
sweepings and construction and demolition waste, and it occupies about one-third of the
total municipal solid waste in landfill sites (Joshi & Ahmed, 2016).
In India, solid waste differs greatly with regard to its composition and hazardous nature,
when compared to MSW in western countries (Sharholy et al., 2008). Solid waste consists
of 30–60% organic matter, 3–10% recyclables, and 30–40% inert matter (Kumar et al. 2009;
Sharholy et al., 2008; Singh et al., 2011; Zhu et al., 2008). There is also a marked distinc-
tion between the solid waste from urban and rural areas. Rural waste is largely agricultural
in nature, but rural areas suffer from ‘pollution sinks’ from the encroaching urban sprawl
(EBTC, 2014). Sharholy et al. (2008) pointed out that the physical and chemical character-
istics of solid waste changes with population density, and indicates the effect of urbaniza-
tion and development. The relative percentage of organic waste in solid waste generally
increases with decreasing socio-economic status; so rural households generate more organic
waste than urban households (Sharholy et al., 2008). The per capita generation rate is high
in some states, such as Gujrat, Delhi and Tamil Nadu, and cities, such as Chennai, Kanpur,
Lucknow and Ahmedabad. On the contrary, it is low in states, such as Meghalaya, Assam,
Manipur and Tripura, and cities, such as Nagpur, Pune and Indore (Sharholy et al., 2008).
Population explosion, coupled with changing lifestyles and improved living standards, has
resulted in an increase in the generation of solid waste. The two major current waste man-
agement challenges are an increasing amount of waste, and the changing composition of
the waste stream.
2.2 Solid waste management systems in india

SWM systems comprise of the following components: storage; collection-transportation;
processing; and disposal. A SWM system focuses on strategies to reduce environmental
impacts and generate value from the waste flow. Figure 1 shows the SWM practices in India.
Wastes generated from process-activities at any house are accounted for through practices,
such as animal feeding, burning, on-site dumping and sales to recyclers. The waste not
accounted for through the above practices are dumped in streets or ditches, or moved to
municipal bins. Commercial waste also flows in a similar manner except for the burying
of waste. The waste collected in municipal bins is transported to waste treatment plants or
landfill sites managed by civic bodies. Ragpickers affect the material flow of waste streams
in terms of segregation and recovery of valuable materials from open dumping sites, com-
munity storage bins or from municipal landfill sites. The material recovered from the rag
pickers or from the waste treatment plants then reaches the recyclers (Section 2.2.1). The
waste processing and treatment stabilizes the waste materials, so that the final residues pro-
duced are not harmful, and incapable of further change or able to find ready entry into
the various natural biogeochemical cycles (Section 2.2.2). Waste management is undergoing
drastic change to offer more options that are sustainable. A short review of recent literature
focusing on the operational dimension of SWM in India is presented in this study.
206

Figure 1. Schematic of solid waste management practice in India.
2.2.1 Waste logistics

First level collection, initial segregation and recycling is undertaken by domestic servants, gar-
bage collectors, rag pickers and waste merchants. Second level collection is completed by non-
governmental organizations (NGO), private contractors and welfare associations (Joshi &
Ahmed, 2016; Nandy et al., 2015; Sharholy et al., 2008). Modes of storage include storage
bins, disposal centers and open collection points. The modes of transport to the transfer sta-
tion include bullock drawn carts, tractor-trailers, tricycles, wheelbarrows, hand rickshaws,
compactors, trucks and dumpers/hydraulic vehicles (Joshi & Ahmed, 2016). Collection and
transportation activities constitute approximately 80–95% of the total budget of the munici-
pal SWM system (Sharholy et al., 2008). Interestingly, collection efficiency is higher in the cit-
ies and states where private contractors and NGOs are employed for this role (Sharholy et al.,
2008). The solid waste flow complexity arises from different sources, waste composition and
flow networks which makes waste management very difficult. Waste segregation in the waste
storage phase, unregulated waste flow in the waste handling phase (Figure 1) and incompat-
ible and outdated transportation technologies, result in poor waste logistics systems.
2.2.2 Waste processing and disposal

MSW processing and disposal practices now include a sophisticated range of options, such
as reuse, recycling, incineration with energy recovery, advanced landfill design and a range
of alternative technologies. In India, the waste treatment practices include aerobic com-
posting in cities, such as Mumbai, Delhi, Bangalore, Ahmedabad, Hyderabad, Bhopal,
Luknow, Vijaywada and Gwalior; vermicomposting in cities, such as Hyderabad, Bangalore,
Mumbai and Faridabad; anaerobic digestion systems in Delhi, Bangalore and Lucknow;
small incinerators for the burning of hospital waste in various cities; gasification system in
Thiruvananthapuram, Nohar and New Delhi; and refuse-derived fuels in Hyderabad, Guntur,
and Vijayawada (Agarwal et al., 2015; Joshi & Ahmed, 2016; Kalyani & Pandey, 2014; Sharholy
et al., 2008). These have all produced varying degrees of success in terms of economic viability
and environmental impact. The nature of the technology for processing is region specific and
depends on the waste characteristics. The energy recovery method appears to be the most pre-
ferred and valued process, but there is a need to understand and use the learning experiences for
generating drivers for technology development and adoption at a macro level.
207

Open dumping and landfills are both practiced. Unscientific open dumps and landfills
are however, creating problems to public health and the environment (Agarwal et al., 2015;
CPCB, 2013; Gupta et al., 2015; Joshi & Ahmed, 2016; Sharholy et al., 2008). Most of India’s
cities/towns are facing problems in the identification of sites for the construction of more
sanitary landfills due to public resistance, rapid growth of urban areas, escalating land prices
and not having a good master plan.
2.2.3 Recycling
Ragpickers affect the material flow of the waste streams in terms of segregation and recovery
of valuable materials from open dumping sites, community storage bins or from municipal
landfill sites. The material recovery from the rag pickers or from waste treatment plants then
reaches the recyclers. The material recovered moves into subsequent production cycle. Recy-
clable materials recovered include paper, plastic, glass, metals, and e-waste (Joshi & Ahmed,
2016; Nandy et al., 2015; Sharholy et al., 2008). Rag pickers, scrap dealers, waste traders and
recycling plants are the elements involved in recycling, and recycling points include houses,
open dumps, bus/train stations and municipal landfills. With respect to the value generated
from the recycling process, Nandy et al. (2015) and Sharholy et al. (2008) take a positive view
while Gupta et al. (2015) takes a negative view.
2.2.4 Organizational framework

Solid waste management in India is a part of public health and sanitation, and is an obliga-
tory function of the municipal bodies. At the national policy level, a number of ministries,
such as the Ministry of Environment and Forests, Ministry of Urban Development, Ministry
of Health and Family Welfare, Ministry of Non-Conventional Energy Sources and Ministry
of Agriculture, are responsible for different aspects of solid waste management (Joseph et al.,
2012). The Central Pollution Control Board and the State Pollution Control Boards form
the administrative core of the SWM sector. Municipal, industrial, biomedical and electronic
waste are governed by different laws and policies: the Municipal Solid Waste Rules 2000;
the Plastics Manufacture and Usage Rules 2003; the Biomedical Waste Rules 1998; and the
Hazardous Wastes Amendment Rules 2000 (Agarwal et al., 2015), for example.
Within the above organizational framework there are four major sectors that operate the
SWM system: the public sector; the private-formal sector; the private-informal sector; and
the community representatives. The public sector comprises of local authorities and local
public departments at city level. The private-formal sector constitutes are large and small
registered enterprises doing collection, transport, treatment, and disposal and recycling. The
private-informal sector consists of small-scale, non-recognized private sectors which com-
prise of waste pickers, itinerant waste buyers, traders and non-registered small-scale enter-
prises. The community representatives are in the form of NGOs.
2.3 Decision-making
SWM systems aim for overall waste management which is the best environmentally, eco-
nomically sustainable for a particular region, and socially acceptable (Agarwal et al., 2015;
Guerrero et al., 2013). A balanced coordination among different factors (social, institutional,
environmental, financial and technical) is needed to achieve an optimal waste management
plan (Guerrero et al., 2013; Srivastava et al., 2014). Decision makers need to be well informed
when developing integrated waste management strategies that are adapted to the needs of a
city, and consider the ability of citizens to pay for the services (Guerrero et al., 2013). The
concept of sustainable waste management is gaining more focus and there is emphasis on
multi-criteria decision analysis.
Decisions pertaining to SWM systems are complex as multiple criteria and multi-actors
are involved. Facility location problems for transfer stations or treatment plant locations
are solved using geographical information systems, mixed integer linear programs, analytic
hierarchy process (ANP) and Technique for Order of Preference by Similarity to Ideal Solu-
tion (TOPSIS) techniques (Choudhary & Shankar, 2012; Khan & Samadder, 2015; Sumathi
208

et al., 2008; Yadav et al., 2016). Multi-criteria decision-making techniques, such as analytic
network process (ANP) and TOPSIS, are used for the selection of alternatives pertaining to
disposal methods, power generation technologies and landfill sites (Khan & Faisal, 2008;
Kharat et al., 2016; Nixon et al., 2013; Thampi & Rao, 2015). Operation research techniques
for a waste management strategy include linear programming, advanced locality models,
goal programming, integer programming, neuro-fuzzy inference systems, artificial neural
networks models, and mathematical models (Phillips & Mondal, 2014; Rathi, 2007; Singh &
Satija, 2015). Operation research techniques can help decision makers to achieve cost savings
as well as to improve waste recovery.
Environmental impact assessments are a widely applied tool used in the selection of a
disposal site or processing strategies that account for ecological, social, cultural and economi-
cal factors. Life cycle assessments have also been widely used to evaluate the environmental
benefits and drawbacks of waste management (Kiddee et al., 2013). Economic evaluations
(Yadhav et al., 2010), full cost accounting (Debnath & Bose, 2014), regression (Parthan et al.,
2012), Strengths-Weaknesses-Opportunities-Threats analysis (Srivastava, 2005), and cost-
benefit analysis (Pandyaswargo & Premkumara, 2014) are a few other generic techniques
that are used for analysis, selection of plans or assessment of services.
Initiatives by the government, NGOs, private companies and the local public toward
sustainable management practices has increased significantly in recent years, and there is a
growing emphasis on community participation in decision-making and operational aspects
of a SWM system (Joseph et al., 2012; Srivastava et al., 2005; Srivastava et al., 2015). Not-
withstanding the recent initiatives, there are numerous operational and sustainability issues
in the working of SWM systems. Primary causes of inefficient systems include lack of plan-
ning, coordinated efforts, leadership, financial constraints, technological issues, institutional
problems within the departments, fragile links with networking agencies, poor public aware-
ness, lack of trained staff and lack of community participation. Such issues arise due to poor
alignment between requirements, constraints and operations in the SWM system.
An integrated and sustainable SWM system design needs to account for public awareness
on waste, environmental impact, value generated from waste processing and disposal, cost-
benefits, social acceptance, and responsiveness of the operational component (Figure 1).
The basic design requirements should be: able to handle a variety of waste; environmentally
friendly; and economically and socially acceptable. This calls for research and developments
in: (i) logistic systems that can handle a variety of waste and are responsive to the public’s
operational demands; (ii) waste processing and disposal technologies that are economical
and environmentally friendly; and (iii) robust decision-making frameworks that enable public
participation.
3 CONCLUSION
This paper provides an insight into the SWM systems and practices in India, and the chal-
lenges faced. The review focused on three major areas: solid waste characteristics; SWM
systems; and decision-making. Based on the literature review, it is evident that SWM is a
complex and challenging problem. The two major waste management challenges are the
increasing amount of waste and the changing composition of the waste stream. Waste seg-
regation, unregulated waste transportation, and incompatible and outdated transportation
technologies result in a poor waste logistics system. There is a need to understand the opera-
tional issues in waste processing and disposal so that better technological development and
adoption, at a macro level, is made possible.
Decision-making processes pertaining to SWM systems are complex because multiple cri-
teria are involved. Invariably, the decision-making practices and solutions generated in India
are unscientific or based on piecemeal approaches. But recent literature provides evidence in
the use of multi-criteria decision-making techniques. The poor performance of SWM systems
are primarily due to poorly aligned requirements, and constraints and operations. A scientific
SWM design is needed which meets the fundamental requirements of being able to handle
209

a variety of waste, be environmentally friendly, and be economically and socially acceptable.
A robust integrated SWM system design places emphasis on active public participation in
decision-making, and operational integration of subcomponents of a SWM system.
REFERENCES
Agarwal, R., Chaudhary, M. & Singh, J. (2015). Waste management initiatives in India for human well
being. European Scientific Journal, 11(10).
Central Pollution Control Board (CPCB, 2013). Status report on municipal solid waste management.
Retrieved from http://www.cpcb.nic.in/divisionsofheadoffice/pcp/MSW_Report.pdfhttp://pratham.
org/images/paper_on_ragpickers.pdf.
Choudhary, D. & Shankar, R. (2012). An STEEP-fuzzy AHP-TOPSIS framework for evaluation and
selection of thermal power plant location: A case study from India. Energy, 42(1), 510–521.
Debnath, S. & Bose, S.K. (2014). Exploring full cost accounting approach to evaluate cost of MSW
services in India. Resources, Conservation and Recycling, 83, 87–95.
European Business and Technology Centre (EBTC, 2014). Snapshot: Waste management in India.
Retrieved from http://ebtc.eu/index.php/knowledge-centre/publications/environment-publications/
174-sector-snapshots-environment/255-waste-management-in-india-a-snapshot.
Guerrero, L.A., Maas, G. & Hogland, W. (2013). Solid waste management challenges for cities in devel-
oping countries. Waste Management, 33(1), 220–232.
Gupta, N., Yadav, K.K. & Kumar, V. (2015). A review on current status of municipal solid waste man-
agement in India. Journal of Environmental Sciences, 37, 206–217.
Joseph, K., Rajendiran, S., Senthilnathan, R. & Rakesh, M. (2012). Integrated approach to solid waste
management in Chennai: An Indian metro city. Journal of Material Cycles and Waste Manage-
ment, 14(2), 75–84.
Joshi, R. & Ahmed, S. (2016). Status and challenges of municipal solid waste management in India: A
review. Cogent Environmental Science, 2(1), 1139434.
Kalyani, K.A. & Pandey, K.K. (2014). Waste to energy status in India: A short review. Renewable and
Sustainable Energy Reviews, 31, 113–120.
Khan, S. & Faisal, M.N. (2008). An analytic network process model for municipal solid waste disposal
options. Waste Management, 28(9), 1500–1508.
Khan, D. & Samadder, S.R. (2015). A simplified multi-criteria evaluation model for landfill site ranking
and selection based on AHP and GIS. Journal of Environmental Engineering and Landscape Manage-
ment, 23(4), 267–278.
Kharat, M.G., Raut, R.D., Kamble, S.S. & Kamble, S.J. (2016). The application of Delphi and AHP
method in environmentally conscious solid waste treatment and disposal technology selection. Man-
agement of Environmental Quality: An International Journal, 27(4), 427–440.
Kiddee, P., Naidu, R. & Wong, M.H. (2013). Electronic waste management approaches: An overview.
Waste Management, 33(5), 1237–1250.
Kumar, S., Bhattacharyya, J.K., Vaidya, A.N., Chakrabarti, T., Devotta, S. & Akolkar, A.B. (2009).
Assessment of the status of municipal solid waste management in metro cities, state capitals, class I
cities, and class II towns in India: An insight. Waste Management, 29(2), 883–895.
Nandy, B., Sharma, G., Garg, S., Kumari, S., George, T., Sunanda, Y. & Sinha, B. (2015). Recovery of
consumer waste in India—A mass flow analysis for paper, plastic and glass and the contribution of
households and the informal sector. Resources, Conservation and Recycling, 101, 167–181.
Nixon, J.D., Dey, P.K., Ghosh, S.K. & Davies, P.A. (2013). Evaluation of options for energy recovery
from municipal solid waste in India using the hierarchical analytical network process. Energy, 59,
215–223.
Pandyaswargo, A.H. & Premakumara, D.G.J. (2014). Financial sustainability of modern composting:
The economically optimal scale for municipal waste composting plant in developing Asia. Interna-
tional Journal of Recycling of Organic Waste in Agriculture, 3(3), 1–14.
Parthan, S.R., Milke, M.W., Wilson, D.C. & Cocks, J.H. (2012). Cost function analysis for solid waste
management: A developing country experience. Waste Management & Research, 30(5), 485–491.
Phillips, J., & Mondal, M. K. (2014). Determining the sustainability of options for municipal solid waste
disposal in Varanasi, India. Sustainable Cities and Society, 10, 11–21.
Rathi, S. (2007). Optimization model for integrated municipal solid waste management in Mumbai,
India. Environment and Development Economics, 12(1), 105–121.
Sharholy, M., Ahmad, K., Mahmood, G. & Trivedi, R.C. (2008). Municipal solid waste management in
Indian cities—A review. Waste Management, 28(2), 459–467.
210

Singh, D. & Satija, A. (2015). Optimization models for solid waste management in Indian cities: A study.
In K.N. Das, K. Deep, M. Pant, J.C. Bansal, A. Nagar (Eds.), Proceedings of Fourth International
Conference on Soft Computing for Problem Solving (pp. 361–371). New Dehli, India: Springer.
Singh, R.P., Tyagi, V.V., Allen, T., Ibrahim, M.H. & Kothari, R. (2011). An overview for exploring the
possibilities of energy generation from municipal solid waste (MSW) in Indian scenario. Renewable
and Sustainable Energy Reviews, 15(9), 4797–4808.
Srivastava, V., Ismail, S.A., Singh, P. & Singh, R.P. (2014). Urban solid waste management in the devel-
oping world with emphasis on India: Challenges and opportunities. Reviews in Environmental Science
and Bio/Technology, 14(2), 317–337.
Srivastava, V., Ismail, S.A., Singh, P. & Singh, R.P. (2015). Urban solid waste management in the devel-
oping world with emphasis on India: Challenges and opportunities. Reviews in Environmental Science
and Bio/Technology, 14(2), 317–337.
Srivastava, P.K., Kulshreshtha, K., Mohanty, C.S., Pushpangadan, P. & Singh, A. (2005). Stakeholder-
based SWOT analysis for successful municipal solid waste management in Lucknow, India. Waste
Management, 25(5), 531–537.
Sumathi, V.R., Natesan, U. & Sarkar, C. (2008). GIS-based approach for optimized siting of municipal
solid waste landfill. Waste Management, 28(11), 2146–2160.
Thampi, A. & Rao, B. (2015). Application of multi-criteria decision making tools for technology choice
in treatment and disposal of municipal solid waste for local self government bodies—A case study of
Kerala, India. The Journal of Solid Waste Technology and Management, 41(1), 84–95.
Yadav, I.C., Devi, N.L., Singh, S. & Devi Prasad, A.G. (2010). Evaluating financial aspects of municipal
solid waste management in Mysore City, India. International Journal of Environmental Technology
and Management, 13(3–4), 302–310.
Yadav, V., Karmakar, S., Dikshit, A.K. & Vanjari, S. (2016). A feasibility study for the locations of
waste transfer stations in urban centers: A case study on the city of Nashik, India. Journal of Cleaner
Production, 126, 191–205.
Zhu, D., Asnani, P.U., Zurbrügg, C., Anapolsky, S. & Mani, S. (2008). Improving municipal solid waste
management in India: A sourcebook for policymakers and practitioners. Washington, DC: World Bank
Publications.
211

Powering India’s villages sustainably: A case study of Bihar
D. Kamath & A. Anctil

Department of Civil and Environmental Engineering, Michigan State University, East Lansing,
Michigan, USA
ABSTRACT: Since independence, the Indian government has been trying to electrify all
rural areas—a daunting task. Bihar, with less than 50% of households electrified, has ambi-
tious plans for increased solar power use. This study compared the environmental and eco-
nomic benefits of centralized and decentralized solar power options to electrify Bihar’s rural
households. A centralized scenario with utility-scale, photovoltaic plants was compared with
decentralized residential rooftop photovoltaic systems. A comparative environmental and cost
life cycle assessment was conducted with a functional unit of 1 kWh electricity to a rural
household in Bihar. The centralized scenario had lower environmental impacts and costs.
However, Bihar’s electricity consumption is mainly residential, which could lead to unutilized
electricity. Considering this made the centralized scenario the worse option. This study tried to
understand the effect of electricity consumption profiles on a system’s environmental impacts
and costs and the role it plays in policy decisions regarding generation capacity increases.
Keywords: Rural electrification, environmental benefits, Life Cycle Assessment, centralized

and decentralized power systems, photovoltaics, rooftop solar, policy, Sustainable Develop-
ment Goals
1 INTRODUCTION
One of the main goals of the Indian Government, since independence, has been to provide
electricity to all its households, especially rural ones. At the village level, six of the 31 Indian
states have 100% electrification. However, some other states are far behind. One such state is
Bihar. With around 47% of households electrified, the state is planning to increase its gen-
eration capacity and transmission and distribution infrastructure (Open Government Data
Platform of India, 2017). The electrification process has been slow, leading to gaps in supply,
which are being met by mushrooming generator businesses (Oda, 2012).
At present, India is trying to increase the renewable fraction of its energy mix and is aggres-
sively pursuing solar power. The National Solar Mission has a revised aim of deploying 100,000
MW of grid-connected solar power by 2022 (Press Trust of India, 2015). Bihar has also aimed
at increasing its installed capacity and is looking forward to solar as its main option (Verma,
2017). The Bihar Government is looking at centralized, utility-scale solar power plants based
on photovoltaic (PV) technology as an answer. Two locations, Kajra and Pirpainti, which had
been chosen for thermal power plants are now being considered for PV power plants (Verma,
2017). At the same time, India is also trying to increase rooftop PV installations by providing
incentives (Ministry of New and Renewable Energy, 2017c). Many studies have compared
decentralized PV systems to other methods of power generation, finding them superior eco-
nomically and environmentally (Gmünder et al., 2010; Molyneaux et al, 2016). A comparison
between centralized and decentralized PV systems has not been conducted.
The present study compared solar PV installations in centralized, utility-scale and decen-
tralized, rooftop scenarios to provide electricity for Bihar’s rural households. The total capac-
ity was assumed to be 400 MWp, similar to what might be installed at Kajra and Pirpainti.
The two scenarios were compared using Life Cycle Assessment (LCA) to understand the
213

environmental impacts and the Levelized Cost of Electricity (LCOE) to understand the
economics.
2 METHODS
2.1 Rural electricity use

Two cases were considered for this study. Case 1 provides electricity to the rural households-
for a restricted time frame (18:00–06:00) and Case 2 for 24 hours. Two synthetic electricity
demand curves were developed with a bottom-up approach, with appliance use as shown in
Table 1, based on a methodology in a previous study (Blum et al., 2013). Assuming a four-
member house hold, this would be on par with Bihar’s per capita consumption of 203 kWh
in 2014–2015 and 258 kWh in 2015–2016 (Tripathi, 2017).
2.2 System design

For rural electrification using solar power, this study considered centralized and decentralized
scenarios, modeled using Homer Pro software (Homer Energy, 2017). A deployment of a total
of 400 MWp was modeled. The solar Global Horizontal Irradiance data for the locations were
obtained from the US National Renewable Energy Laboratory’s National Solar Radiation
Database (National Renewable Energy Laboratory, n.d.). The temperature data were obtained
from the NASA Surface Meteorology and Solar Energy Database (NASA, n.d.). Subsequently,
17.5% efficiency multi-Si PV panels with a lifetime of 30 years, along with a 98% efficiency
inverter, which has a lifetime of 15 years (Fu et al., 2017), were used as inputs. The PV panels
were assumed to be latitude tilted. Since the roof angle in India usually lies between 10–30°
(B.I.S. IS 875–1987), latitude tilting is easily obtained from the residential rooftops.
The centralized scenario had two PV power plants located in Kajra and Pirpainti, Bihar,
selected because of the plans to set up thermal power plants here, which fell through. The
land procured for these projects was approximately 1000 acres each (Verma, 2017), provid-
ing enough area for a 200 MWp capacity solar power plant at both locations (International
Finance Corporation, 2015). Both PV plants were assumed to be connected to the existing
Table 1. Appliance use considered for Bihar’s rural households (HH).
Case Case 1: 0.92 kWh/day Case 2: 1.88 kWh/day
Usage Usage
Power Quantity duration Quantity duration
consumption per HH per day Season per HH per day Season
Appliances W No. NA NA No. NA NA
Light bulb 16 2 18:00–00:00 All year 2 18:00–00:00 All year

indoor
Light bulb 16 1 18:00–06:00 All year 1 18:00–06:00 All year
outdoor
TV 19″ 80 0.2 (1 every 18:00–23:00 All year 1 18:00–23:00 All year
5 HH)
Heater 400 0.2 (1 every 23:00–0:00, Winter 0.6 (3 every 23:00–0:00, Winter
5 HH) 4:00–05:00 5 HH) 4:00–05:00
Cooler 300 Summer 0.2 (1 every 22:00–23:00 Summer
2 HH)
Fan 25 0.5 (1 every 22:00–06:00 Summer 1 20:00–21:00, Summer
2 HH) 00:00–06:00
Refrigerator 100 NA NA NA 4 per 30 HH 17:00–09:00 All year
Street light 80 0.2 18:00–06:00 All year 1 18:00–06:00 All year
214

electricity grid and included a transmission and distribution system, based on the planned
transmission and grid system extension in preparation for the thirteenth development plan
(Central Electricity Authority, 2016). The energy losses were assumed to be 20% based on
various reports (Central Electricity Authority, 2017; The World Bank, n.d.).
The decentralized scenario hadhousehold-level rooftop solar systems with PV panels and
lead-acid batteries as storage, because of the reasonable costs (Ministry of New and Renew-
able Energy, 2017). Loma (Bhijrauli) was taken to be a proxy location to obtain the solar
insolation. This system was designed by optimizing for net present cost. The costs of the
PV panels, inverter and batteries were assumed based on US benchmark data as proxy and
market rates (Fu et al., 2017; Vijayakumar, 2015).
Based on the design requirement for one household, the number of households equaling
400 MWp PV installations was calculated. This demand was assumed to be the first scenar-
io’s demand. The centralized and decentralized scenarios were compared with the assump-
tion that both had a total PV installed capacity of 400 MWp and took care of the demand of
the same number of households.
2.3 Life cycle assessment

LCA was used to compare the greenhouse gas emissions and the embodied energy of the
PV systems using the ReCipe Midpoint (Heirarchist) and the Cumulative Energy Demand
(CED) method. The life cycle inventories for the PV system and the electricity grid were
based on the Ecoinvent 3 database (Wernet et al., 2016). The transmission and distribution
system inventory and the lead-acid battery inventory were based on previous publications
(Jorge et al., 2012a, 2012b; Rydh, 1999).
For this study, the functional unit, which is the reference unit used to compare systems
or products in LCA, is 1 kWh of electricity to a rural household in Bihar. The assessment
excluded transportation and the end-of-life of the system.
2.4 Levelized cost of electricity

The LCOE was used to calculate the average cost of useful electricity produced. It was calcu-
lated by dividing the annual costs incurred during the lifetime of the system (annualized costs)
by the electricity produced per year. Its unit was Rs/kWh, which allowed for an easy compari-
son with conventional electricity costs or tariffs. The annualized costs were based on the net
present costs which were the discounted costs incurred in the lifetime of the system, including
the capital, operation and maintenance, and replacement costs (Homer Energy, 2017).
The nominal discount rate considered was 11.75% (Clean Development Mechanism Exec-
utive Board, 2011) and the expected inflation rate was 4.6% (Organisation for Economic Co-
operation and Development, n.d.). Data for the costs incurred in the lifetime of the system
were obtained from governmental benchmarks ((Deo et al., 2010a, 2010b; Gakkhar, 2017;
Pradhan et al., 2016).
3.1 System design

Homer Pro was used to find the PV generation from the centralized scenario and the amount of
demand that was met by the PV system and the electricity grid. For the decentralized scenario,
Homer Pro was used to optimize the system’s net present cost. Tables 2 and 3 show the results.
The number of households accounting for 400 MWp PV installation was also determined.
For Case 1, 800,000 households were considered as the rooftop PV size was 0.5 kWp,
whereas, for Case 2, the number decreased to 266,667. This caused the total consumption for
Case 1 with 800,000 households to be higher, even though the household consumption was
much lower than that in Case 2.
215

Table 2. Centralized scenario generation for both locations.
Location Kajra Pirpainti Total Kajra Pirpainti Total
PV (MWp) 200 200 400 200 200 400

Converter (MW) 200 200 400 200 200 400
Total use (GWh/year) 134.3 134.3 268.6 91.5 91.5 183
Electricity needed from grid 128.6 128 256.5 83.9 83.7 167.6
(GWh/year)
Electricity sent to grid 331.7 336.5 668.2 340.4 324.7 665.1
(GWh/year)
Table 3. Design for decentralized scenario for one household.
PV (kWp) 0.5 1.5

Converter (kW) 0.25 0.75
Battery capacity (kWh) 5 10
Electricity sold yearly when grid connected 380 1429
(kWh)
No. of HH considered (total installation 800000 266667
of 400 MWp PV)
3.2 Life cycle assessment

A comparative LCA was conducted with the existing electricity grid as the baseline. For each
case, five scenarios were compared for sensitivity analysis. The five scenarios can be described
as follows:
1. Existing electricity grid utilized to meet demand (centralized conventional).
2. Centralized system with all electricity utilized either by households or sent to grid (central-
ized PV).
3. Centralized system with excess electricity left unutilized (centralized PV excess).
4. Decentralized system, off-grid, causing excess electricity to go unutilized (decentralized
off-grid).
5. Decentralized system, on-grid, causing excess electricity to be sent to grid (decentralized
on-grid).
A mismatch between the load and the generation is seen. In the centralized scenario, the
load is met by the existing electricity grid and the excess electricity from the PV generation is
utilized elsewhere (Scenario 2). Scenario 3 is modeled to consider the case when there is no
utilization of the excess electricity. In the decentralized system, the lack of match between
the load and generation is compensated using batteries (Scenario 4). Scenario 5 considers the
utilization of the excess electricity still produced.
The Global Warming Potential (GWP) and CED for all scenarios are shown in Figures
1(a) and 1(b). When the household load was 0.92 kWh/day, as in Case 1, the GWP and CED
were maximum for Scenario 3 (centralized PV excess), higher than Scenario 1 (centralized
conventional). When the excess electricity was utilized, as in Scenario 2, it was one of the
best scenarios, with negative impacts compared to Scenario 1. Scenario 5 offered the most
environmental benefits in this case. However, when the load is 1.88 kWh/day, design changes
led to Scenario 3 having impacts similar to Scenario 1, and utilizing the excess electricity led
to Scenario 2 having the lowest impacts, followed by Scenario 5.
Table 4 describes the scenarios diagrammatically along with their system boundaries, typi-
cal demand and PV generation, showing the mismatch between the load and the generation.
216

Figure 1. Figure shows (a) GWP, (b) CED and (c) comparison of LCOE with the present tariff.
Table 4. System boundaries for each scenario with a comparison of demand and PV generation.
3.3 Levelized cost of electricity

Scenarios in which the excess electricity was not sent to the grid, such as with high electricity
wastage, had higher LCOE, as seen in Figure 1(c). Similar to the LCA results, Scenario 2 had
the lowest LCOE. However, when the excess electricity was not utilized, the costs were much
higher in all the scenarios. Comparing both the LCA and LCOE results, the best scenario for
both cases was the centralized PV system. However, it is important to consider the unutilized
excess electricity, because 66% of Bihar’s total electricity generated is consumed by the resi-
dential sector (Molyneaux et al., 2016). Hence, the peak demand occurs during the evening
when there is no PV electricity generation as shown in Table 4. Bihar is already in a situation
where the power generation follows the load profile closely. PV electricity generation without
any storage could aggravate this situation, leading to a large amount of unutilized excess
electricity, as in Scenario 3 in this study. This situation could be averted with more industrial
and commercial projects or grid storage.
217

Grid storage is picking up in popularity in India recently with a 10 MW grid storage array
planned for Delhi (Maloney, 2017). However, it is still not cost competitive when compared to
PV projects with no storage, leading to the cancellation of tenders for solar farms with storage
(Chandrasekaran, 2017). Until then, decentralized systems, either on- or off-grid, might be a
good option for rural electrification, depending on the completion of transmission and distribu-
tion projects and access to interconnection points, as well as the level of regional development.
4 CONCLUSION
This study compared rural electrification options for Bihar using PV technology, with the
existing grid as the baseline. Overall, the best scenario in terms of GWP, CED and LCOE
was the centralized PV system. However, in Bihar, such a system could lead to significant
wastage of electricity, increasing the GWP, CED and LCOE compared to the off-grid decen-
tralized PV system. This must be considered when comparing rural electrification options as
most of Bihar’s consumption is from the residential sector, with a peak demand at times of
low or no solar insolation.
The electricity consumption profile could impact the system’s environmental impacts and
costs, turning a seemingly best scenario to the worst, because of unutilized generated elec-
tricity. It plays a crucial role in policy decisions and should be considered when looking at
capacity increases for energy generation.
Previous studies have also cited high connection costs leading to low connection rates
(Cook, 2011), which is the case when there is no correlation between the consumption and
generation profiles. Electricity would be unaffordable and hence, inaccessible to the poor.
However, the cost might be reduced by increasing industrial development in the region.
Future studies should be aimed to understand the effect of such a development in the area
and if the benefits of a rural electrification program would reach the intended population,
and what policies would aid this endeavor.
REFERENCES
B.I.S. IS 875–1987, “Code of Practice for design loads (other than earthquake) for building and struc-
ture”, Part, 3. New Delhi.
Blum, N.U., Sryantoro Wakeling, R., & Schmidt, T.S. (2013). Rural electrification through village
grids—Assessing the cost competitiveness of isolated renewable energy technologies in Indonesia.
Renewable and Sustainable Energy Reviews, 22, 482–496.
Central Electricity Authority. (2016). Bihar: Transmission System Study for 13th Plan. New Delhi.
Central Electricity Authority. (2017). Power Sector: Executive Summary for the Month of Jan, 2017.
New Delhi.
Chandrasekaran, K. (2017, July 20). Crashing solar tariffs crush storage plans. The Economic Times.
New Delhi.
Clean Development Mechanism Executive Board. (2011). Guidelines on the Assessment of Investment
Analysis. Bonn.
Cook, P. (2011). Infrastructure, rural electrification & development. Energy for Sustainable Develop-
ment, 15(3), 304–313.
Deo, P., Jayaraman, S., Verma, V.S., & Dayalan, M.D. (2010a). Benchmark Capital Cost for 400/765 kV
Transmission Lines. New Delhi.
Deo, P., Jayaraman, S., Verma, V.S., & Dayalan, M.D. (2010b). Benchmark Capital Cost for Substation
associated with 400/765 kV Transmission System. New Delhi.
Fu, R., Feldman, D., Margolis, R., Woodhouse, M., & Ardani, K. (2017). U.S. Solar Photovoltaic Sys-
tem Cost Benchmark : Q1 2017. National Renewable Energy Laboratory. Golden, CO.
Gakkhar, N. (2017). Benchmark Cost for “Grid Connected Rooftop and Small Solar Power Plants Pro-
gramme” for the year 2017−18. New Delhi.
Gmünder, S.M., Zah, R., Bhatacharjee, S., Classen, M., Mukherjee, P., & Widmer, R. (2010). Life cycle
assessment of village electrification based on straight jatropha oil in Chhattisgarh, India. Biomass
and Bioenergy, 34(3), 347–355.
218

Homer Energy. (2017). Levelized cost of electricity. Homer Pro Support Documentation. Boulder,
Colorado: Homer Energy.
International Finance Corporation. (2015). Utility-Scale Solar Photovoltaic Power Plants: A Project
Developer’s Guide. Washington, D.C.
Jorge, R.S., Hawkins, T.R., & Hertwich, E.G. (2012a). Life cycle assessment of electricity transmission
and distribution—part 1: power lines and cables. The International Journal of Life Cycle Assessment,
17(1), 9–15.
Jorge, R.S., Hawkins, T.R., & Hertwich, E.G. (2012b). Life cycle assessment of electricity transmission
and distribution—part 2: transformers and substation equipment. The International Journal of Life
Cycle Assessment, 17(2), 184–191.
Maloney, P. (2017). AES and Mitsubishi pair up for India’s first grid-scale storage project. Utility Drive.
New York.
Ministry of New and Renewable Energy. (2017). National Solar Mission—An Appraisal. New Delhi:
Lok Sabha Secretariat.
Molyneaux, L., Wagner, L., & Foster, J. (2016). Rural electrification in India: Galilee Basin coal versus
decentralised renewable energy micro grids. Renewable Energy, 89, 422–436.
NASA. (n.d.). Surface Meteorology and Solar Energy Database. Hampton, VA.
National Renewable Energy Laboratory. (n.d.). National Solar Radiation Data Base 1991–2005 Update:
Typical Meteorological Year 3. Lakewood, CO.
Oda, H. (2012). Progress and Issues in Rural Electrification in Bihar, India: A Preliminary Analysis. IDE
Discussion Papers. Chiba, Japan.
Open Government Data Platform of India. (2017). State-wise Village Electrification state as on date.
New Delhi.
Organisation for Economic Co-operation and Development. (2017). Inflation Forecast (indicator).
OECD Data.
Pradhan, G.B., Singhal, A.K., Bakshi, A.S., & Iyer, M.K. (2016). Determination of Benchmark Capi-
tal Cost Norm for Solar PV power projects and Solar Thermal power projects applicable during FY
2016–17. New Delhi.
Press Trust of India. (2015, June 17). Solar Power Target Reset to One Lakh MW. Business Standard.
New Delhi.
Rydh, C.J. (1999). Environmental assessment of vanadium redox and lead-acid batteries for stationary
energy storage. Journal of Power Sources, 80(1–2), 21–29.
The World Bank. (2017). Electric Power Transmission and Distribution Losses (% of output).
Tripathi, P. (2017, February 25). No Additional power capacity to Bihar by 2018–19. The Economic
Times. Patna, Bihar.
Verma, S.K. (2017, August 6). Solar power plants on Nitish radar. The Telegraph. Patna, Bihar.
Vijayakumar, S. (2015, May 10). Scaling the Powerwall. The Hindu. Chennai, Tamil Nadu.
Wernet, G., Bauer, C., Steubing, B., Reinhard, J., Moreno-Ruiz, E., & Weidema, B. (2016). The ecoin-
vent database version 3 (part I): overview and methodology. International Journal of Life Cycle
Assessment, 21(9), 1218–1230.
219

Geomechanics and foundation engineering

Role of sodium silicate in strength development of cement treated

clayey soil admixed with composite promoter
K. Raj Keerthi & J. Bindu

Department of Civil Engineering, College of Engineering, Thiruvananthapuram, Kerala, India
ABSTRACT: Friendly sodium silicate and promoters, which are compatible with cement
are used to obtain improved soil properties. The possibility of using cement and sodium sili-
cate admixed with composite promoters to improve the strength of soft clay was analysed in
the present study. The influential factors involved in this study are the proportion of sodium
silicate binding agent and the curing time. The unconfined compressive strength of stabilized
clay at different ages is tested. Based on literature study, the selected composite promoters for
the present study comprises of CaCl2 & NaOH. For the ordinary Portland cement (OPC) and
sodium silicate admixed with composite promoters system, the permeation of the CaCl2 and
NaOH solutions is expected to facilitate the precipitation of Ca(OH)2 at a molar ratio 1:1
and found significantly improves strength of soft clay. More importantly, it is found that the
selected clay stabilizer in much less dosage is needed to achieve the equivalent improvement in
strength compared with cement, hence can be a more effective and ecofriendly clay stabilizer.
1 INTRODUCTION
Construction of buildings and other civil engineering structures on weak or soft soil is highly
risky because such soil is susceptible to differential settlements due to its poor shear strength
and high compressibility. Soil stabilization is the process of improving the physical and
engineering properties of problematic soils to some predetermined targets. Sodium silicates
have been widely used as supplementary cementing materials substituting ordinary Portland
cement to improve the soil properties. OPC is the most commonly used stabilizer since it is
readily available at reasonable cost. Nevertheless, a major issue with using OPC is that its
production processes are energy intensive and emit a large quantity of CO2. To improve the
environmental acceptability and to reduce the construction cost of the deep mixing method,
the partial replacement of the cement by supplementary cementing materials such as sodium
silicate is one of the best alternative ways.
Application of sodium silicate for geotechnical works has been reported by many research-
ers. Used as a component of soil stabilizer, sodium silicate has unique advantages:
(i) reliable and proven performance, (ii) safety and convenient for construction, and
(iii) environmental acceptability and compatibility (Rowles and O’Connor, 2003; Ma et al.,
2014). In order to investigate the possibility of using cement and sodium silicate admixed
with composite promoters to improve the strength of soft clayey soil, Thonnakkal soft clay
is considered. These deposits are composed of silty clay, having extremely low shear strength
and high compressibility.
The aim of this present study is to achieve an OPC-based clay stabilizer which has the
equivalent enhancement of the mechanical properties as a higher content of OPC. The effect
of sodium silicate on the strength development of samples stabilized with OPC and compos-
ite promoter was investigated. The unconfined compressive strength was used as a practical
indicator to investigate the strength development. The binders consisting of OPC, sodium
silicate, and composite promoters. The present study aimed to obtain an optimum dosage of
223

sodium silicate and to study its effect on strength development upon curing, since there are
conflicts exist on optimum dosage of sodium silicate additive. The use of single promoters
used as the component of clay stabilizers have several disadvantages, restrict the application
of single promoters in the OPC and sodium silicate system. The composite promoters, com-
prised of CaCl2/Ca(OH)2 or Ca(OH)2/NaOH, have no advantage in terms of strength devel-
opment in comparison with the same addition of single Ca(OH)2 or NaOH. The selected
composite promoter (CN) for the present study consists of CaCl2 and NaOH at the mass
ratio of 1:1, gives better strength improvement (Cong Ma et al., 2015).
2 MATERIALS
2.1 Clay soil

The samples were collected from Thonnakkal region in the Trivandrum district of Kerala.
Index properties of the soils are summarized in Table 1.
2.2 Cement
A 43 grade ordinary Portland cement (OPC) was used in this study. Properties are listed in
Table 2.
2.3 Sodium silicate

Sodium silicate (SS), in powdered form is used. It consists of SiO2 (29.48%) and Na2O
(9.52%).
2.4 Composite promoter

Composite promoter of CaCl2 & NaOH in the molar ratio 1:1 is used for the present study.
Sodium hydroxide (NaOH), a flaked solid at room temperature. Calcium chloride (CaCl2), an
anhydrous powder, was used as an accelerator. It can serve as a source of calcium ions in the
solution. Unlike many other calcium compounds, calcium chloride is soluble. These powdered
promoters are all chemically pure and are obtained from Laboratory suppliers Trivandrum.
3 METHODOLOGY
3.1 Sample preparation

The samples were prepared by varying dosages of sodium silicate (0.5%, 1%, 1.5%, 2%, 2.5%
and 3%), cement content and composite promoter (CaCl2 & NaOH) dosage were kept constant
throughout the test. Cement dosage is fixed as 10% and composite promoter dosage is fixed
Table 1. Basic properties of clay soil.
Soil property Values Table 2. Properties of OPC.
Liquid limit (wl) (%) 45.2 Properties Values

Plastic limit (wp) (%) 25.6
Plasticity index (Ip) (%) 19.6 Specific surface (m2/kg) 280
Shrinkage limit (ws) (%) 20.8 Initial setting time (min) 180
Specific gravity G 2.5 Final setting time (min) 260
Undrained shear strength, Su (kPa) 124.8 Compressive strength (MPa)
Clay content (%) 55 3 days 37
Silt content (%) 45 7 days 45
IS classification CI 28 days 55
224

as 1% at 1:1 molar ratio for the present study. Samples were prepared at optimum moisture
content and maximum dry density obtained from standard proctor test. For each variations of
sodium silicate dosage, proctor test were conducted in order to obtain the moulding water con-
tent and dry density of samples. Samples were prepared on PVC mould of 4 cm diameter and
length of 8 cm (aspect ratio 1:2). The samples was allowed to cure for a required curing period
in an air tight moisture proof desiccators maintained at a relative humidity of more than 95%.
3.2 Experimental programs

Unconfined compressive strength test (UCS Studies) was conducted to estimate the effect of
sodium silicate on the strength development of samples stabilized with OPC and composite
promoter ( OPC-10% and CN-1%). The UCS tests were carried out according to IS 2720
(part 10):1991. The samples were tested using a compression device fitted with a proving ring
for measuring the applied load. The axial displacements were measured using a dial gauge.
A constant strain rate was adopted in testing the samples. UCS test were conducted on sam-
ples with varying sodium silicate dosage and on varying curing periods.
4.1 Compaction results

The moisture-density (compaction) curves for the cement treated soils were determined using
the method described in ASTM D558-04. Figure 2 shows the variation of the maximum dry
density with the moisture content for the tested samples. Upon addition of sodium silicate
maximum dry density is found to be higher compared with samples treated with cement
alone. Also there is a marked reduction in optimum moisture content, thus showing improve-
ment in soil strength. Best compaction characteristics is shown by samples treated with 1%
Figure 1. UCS experimental set up.
225

Figure 2. Variation of the maximum dry density with the moisture content for the tested samples.
Figure 3. Stress-strain curve for samples treated with different sodium silicate dosage for a curing
period of 14 days.
sodium silicate. A percentage increase of 15.35% in dry density and a percentage decrease of
26.7% in moisture content was observed in samples treated with 1% sodium silicate additive.
4.2 UCS test results

4.2.1 Effect of sodium silicate dosage
The UCS tests were conducted with varying percentages of sodium silicate additive 0.5%,
1%, 1.5%, 2%, 2.5% and 3% and samples were prepared for a cement content of 10% and CN
of 1%. Figure 3 represent UCS test result for varying sodium silicate additive dosages for a
curing period of 14 day. It can be seen that when sodium silicate is used along with cement,
the unconfined compressive strength is higher compared with samples treated with cement
alone. It observed that the stiffness of the cement treated soil increases after sodium silicate
addition and the stiffness continues to increase with higher dosages of sodium silicate. For a
226

particular cement content of 10% and composite promoter (CaCl2 & NaOH) dosage of 1%
at 1:1 molar ratio, the optimum sodium silicate dosage is found to be 1%, where strength
improvement is maximum. Further increase in sodium silicate dosage from 1%, there is
reduction in strength development. Strength development is found to be higher on whole
number fraction dosage of sodium silicate when compared to decimal fraction dosage of
sodium silicate for a given composite promoter (CaCl2 & NaOH) dosage of 1%.
4.2.2 Effect of curing on strength development

Table 3, shows data representing strength development of OPC stabilized clay stabilizer
upon curing periods. Strength development of samples increases up on increasing the curing
Table 3. Variation of unconfined compressive strength for different

dosages of sodium silicate for a curing period of 14 days.
14 day unconfined compressive

Sample proportion strength (kN/m2)
Clay alone 163.3

Clay+10%OPC 1407.2
Clay+10%OPC+0.5%SS+1%CN 1327.6
Clay+10%OPC+1%SS+1%CN 2316.2
Clay+10%OPC+1.5%SS+1%CN 1518.0
Clay+10%OPC+2%SS+1%CN 1873.3
Clay+10%OPC+2.5%SS+1%CN 1377.1
Clay+10%OPC+3%SS+1%CN 2073.7
Table 4. Unconfined compressive strength of different samples for varying curing periods.
Unconfined strength for varying curing days (qu) (kN/m2)
Sample proportion 0th day 3rd day 7th day 14th day
Clay alone 124.1 136.2 155.3 163.3

Clay+10%OPC 262.1 882.5 1155.6 1407.2
Clay+10%OPC+0.5%SS+1%CN 136.9 765.9 1327.6 1507.3
Clay+10%OPC+1%SS+1%CN 410.5 1410.2 2316.2 2555.8
Clay+10%OPC+1.5%SS+1%CN 148.7 1176.8 1518.0 1700.0
Clay+10%OPC+2%SS+1%CN 561.6 1378.5 1873.3 2131.2
Clay+10%OPC+2.5%SS+1%CN 171.1 1082.1 1377.1 1650.6
Clay+10%OPC+3%SS+1%CN 142.2 1260.1 2073.7 2290.8
Figure 4. Effect of curing on strength development for samples treated with varying sodium silicate
dosages.
227

Figure 5b. Failure pattern of sample
Figure 5a. Failure pattern of sample after curing (at 1% SS content).
without curing (at 1% SS content).
period and a significant increase in strength is found after 7 days of curing. More brittle and
a sudden failure was observed with curing.
5 CONCLUSION
The conclusions made based on the present study on clayey soil are:
This paper analysed the strength development in OPC and sodium silicate-stabilized clay
with composite promoter (NaOH & CaCl2). NaOH and CaCl2 as a composite promoter at
mass ratio 1:1 along with addition of sodium silicate on cement stabilized clay significantly
improves the strength
• Strength development is found to be higher on whole number fraction dosage of sodium
silicate when compared to decimal fraction dosage of sodium silicate for a given composite
promoter (CaCl2 & NaOH) dosage of 1%.
• It can be observed that for a particular cement content of 10% and a composite promoter
(CaCl2 & NaOH) at combination of 1:1 molar ratio, 1% sodium silicate was found to be
most effective dosage for strength development and observed that upon increasing sodium
silicate dosage bonding gel related to sodium silicate become weaker.
• Strength development of samples increases upon increasing the curing period and a sig-
nificant increase in strength is found after 7 days of curing.
REFERENCES
Bindu, J., and Ramabhadran, A. (2011). “Study on cement stabilized Kuttanad clay.” Proc., Indian Geo-
technical Conf., Indian Geotechnical Society (IGS), Kochi, India, 465–68.
Cong Maa, Zhaohui Qina, Yingchun Zhuangb, Longzhu Chena, Bing Chena. (2015). “Influence of
sodium silicate and promoters on unconfined compressive strength of Portland cement-stabilized
clay.” Soils and Foundations 2015; 55(5): 1222–232, Elsevier.
Mohammad Vali Vakili, Amin Chegenizadeh, Hamid Nikraz, Mahdi Keramatikerman. (2016). “Inves-
tigation on shear strength of stabilised clay using cement, sodium silicate and slag.” Applied Clay
Science 03768; Elsevier.
Saroglou I. Haralambos. (2013). “Compressive Strength of Soil Improved with Cement.” ascelibrary.
org on 06/06/13.
Sina Kazemian, Arun Prasad, Bujang B.K. Huat, Vahed Ghiasi, Soheil Ghareh (2012). “Effects of
Cement–Sodium Silicate System Grout on Tropical Organic Soils.” Arab J Sci Eng (2012) 37:
2137–2148.
Suganya, K., P.V. Sivapullaiah (2016). “Role of Sodium Silicate Additive in Cement-Treated Kuttanad
Soil.” Journal of Materials in Civil Engineering, ASCE, ISSN 0899-1561.
228

Stabilisation of Kuttanad clay by environment friendly

methods—a review
K. Kannan
Department of Civil Engineering, Marian Engineering College, Trivandrum, India
S. Gayathri
Department of Civil Engineering, John Cox C.S.I Memorial Institute of Technology, Trivandrum, India
Abiya K. Jose, Lakshmi P. Nair & Greeshma Das

Department of Civil Engineering, Marian Engineering College, Trivandrum, India
P. Vinod
Department of Civil Engineering, Government Engineering College, Trichur, India
ABSTRACT: Kuttanad clay is a well known soil group, known for its low shear strength and high
compressibility, making it almost always unusable and inconstructible on in its natural state. The
traditionally used stabilization techniques such as preloading and chemical grouting are unsuit-
able in the present scenario, as they are either outpaced or affect the environment aggressively.
This paper reviews some of the sustainable methods of ground improvement which have been
used and reported in highly plastic clays. These techniques can potentially be used in Kuttanad
clay also, provided further studies are conducted. The paper also reports the effect of one tech-
nique, Microbial Induced Calcite Precipitation (MICP), on the liquid limit of Kuttanad clay.
1 INTRODUCTION
When only poor quality soil is available at the construction site, the best option is to modify
the properties of the soil so that it meets the design requirements. The process of improving
the strength and durability of soil is known as soil stabilization. It is the alteration of soils to
enhance their physical properties. Stabilization can increase the shear strength and control
the shrink-swell properties, thus improving the load bearing capacity of the soil.
Kuttanad is situated in the central half of Kerala covering an area of approximately 1100
sq.km and lies 0.6 m to 2.2 m below the mean sea level. Kuttanad clay is an important soil
group, well known for its low shear strength and high compressibility. Soil in this region is
soft black or grey marine clay composed of minerals such as montmorillonite, kaolinite, iron
oxide and aluminum oxide (Vinod and Bindu, 2010). The natural water contents of this soil
are very high and close to liquid limit, sometimes even exceeding it. The typical kuttanad soil
consists primarily of silt and clay fraction. It is a weak foundation material, with a number of
failures to structures and embankment reported. Since Kuttanad is the rice bowl of Kerala,
any ground improvement technique adopted in this region should be eco-friendly and should
never cause any harm to the environment, especially to the soil and water. Thus, the pres-
ently used methods of physical improvement such as preloading and chemical improvement
by addition of adulterants might prove to be inefficient in the present scenario, wherein the
focus is on fast sustainable technologies. All cement based techniques may seem harmless to
the public eye, but add to carbon footprint heavily during manufacture. The improvement of
subsoil using alternative biological or ecofriendly chemical methods is thus a growing con-
cern, and the focus of the present paper. In particular, the paper focuses on some alternative
sustainable techniques which can be potentially utilized to improve Kuttanad clay.
229

Table 1. Constituents of Bagasse Ash (Kharade et al., 2014).
Constituent Percentage
Silica (SiO2) 64.38

Magnesium Oxide (MgO) 0.85
Calcium Oxide (CaO) 10.26
Iron Oxide (Fe2O3) 4.56
Sodium Oxide (Na2O) 1.05
Pottassium Oxide (K2O) 3.57
Alumina (Al2O3) 11.67
2 BAGASSE ASH
Bagasse is a residue obtained from the burning of bagasse in sugar producing factories.
Bagasse is the cellular fibrous waste product after the extraction of the sugar juice from cane
mills. For each 10 tons of sugarcane crushed, a sugar factory produces nearly 3 tons of wet
bagasse which is a byproduct of the sugar cane industry (Kharade et al., 2014). When this
bagasse is burned the resultant ash is bagasse ash. Bagasse shows the presence of amorphous
silica, which is an indication of pozzolonic properties, responsible in holding the soil grains
together for better shear strength. Pozzolanic material is very rich in the oxides of silica and
alumina and sometimes calcium. Pozzolans usually require the presence of water in order for
silica to combine with calcium hydroxide to form stable calcium silicate which has cementi-
tious properties, which can then develop good bonding between soil grains in case of weak
soil. Table 1 shows the constituents of bagasse ash (Kharade et al., 2014).
2.1 Effects of bagasse ash on clays

The stabilization of soil using bagasse ash in different percentage has been reported. With
the increase of bagasse ash the MDD of mix is reported to be decreasing and CBR value
of the mix is increasing up to an optimum percentage of bagasse ash in mix. As CBR value
increased, the strength characteristics of clayey soil also increased (Kanchan & Jaiwid, 2013).
The results of different tests which are conducted with varying percentage of bagasse ash to
check the effect on swelling pressure and on basic properties shows that the bagasse ash effec-
tively dries wet soils, by producing a significant reduction in water content, and provides an
initial rapid strength gain. It also decreases the swell potential of expansive soils by replacing
some of the volume and by cementing the soil particles together (K.S. Gandhi, 2012). From
the reported tests, it was obtained that the index properties like Atterberg limits and Plastic-
ity Index, and free swell index decreases to a lowest value at 20% bagasse ash for black cotton.
The bagasse ash is added in lower percentage and it gives good results, showing that bagasse
ash is an effective adulterant (K.S. Gandhi, 2014).
3 BIOENZYME
Bioenzyme in soil stabilization have emerged as a revolutionary technique which is

becoming popular worldwide. Bioenzymes are nontoxic, nonflammable substances that
are obtained from plant extracts (Saini and Priyanka, 2015). Bioenzyme reduces the voids
between the particles of soil and minimizes the amount of absorbed water in the soil so
that compaction can be maximum. These enzymes have been proven to be very effective
and economical. The use of bioenzyme in soil stabilization is not very popular, mostly
due to a lack of awareness. Reported literature shows soil types stabilized by the Bio-
enzymes include sandy clay, silty clay, sandy silt, plastic and non-plastic clay, sandy loam,
fine loam, and loam mixed with clay. The dosage levels of Bio-enzymes vary from 1 to
5 litres for 5 m3 of soil depending on the soil type and characteristics. The amount of
230

Table 2. Properties of Terrazyme (Shirsath et al., 2017).
Property Value/specification
Boiling point 212°F

Specific gravity 1 to 1.09
pH value 4.3 to 4.6
Hazardous content None
Appearance Brown colored liquid
Reactivity Data Stable
dilution water depends on in-situ moisture content of soil. There are many bioenzymes
available namely renolith, permazyme, terrazyme, fujibeton etc. Most commonly used one
is terrazyme.
3.1 Terrazyme
Terrazyme modified from vegetable extracts is specially formulated to modify the engineer-
ing properties of soil, usually applied as a mixture with water. Terrzayme acts on the soil by
reducing the voids between the particles minimizing the adsorbed layer of water and maxi-
mizing compaction. It reacts with the organic materials to form cementatious material, bring-
ing about a decrease in the permeability and increase in chemical bond, creating a permanent
structure resistant to weather, wear and tear (Gupta etal., 2017). Table 2 shows the properties
of terrazyme (Shirsath et al., 2017).
3.2 Mechanism of terrazyme

Soil (clay) has surface negative charge particles making the soil attracted to the positive
charges and get neutralized. It is found that the adsorbed layer of water contains metals like
Na, K, Al, Mg which is the reason for the bond between negative clay particles and water
molecules. Thus a significant layer of water is created around soil particles and to attain
compaction it is necessary to eliminate the water layer around the particle.
Terrazyme reduces the dielectric charge in water molecule thus creating pressure on
the positive metal ions to release the free water. This breaks the electrostatic potential
barrier thereby reducing the thickness of the diffused double layer so that the soil parti-
cles come closer and attain greater compaction with less compactive effort (Eujine et al,
2014).
3.3 Effects of terrazyme on clays

Effect of terrazyme is different for different types of soil and mainly depends on dosage
and curing time. With the addition of terrazyme a significant increase in the values of
both soaked and unsoaked CBR has been observed, by as much as 4 times for clayey silts
(Saini and Vaishnava, 2015). This is because of increased compaction creating stronger
bond which resist penetration. The OMC and consistency limits have also been found to
decrease, indicating the denseness of the soil. Permeability of the soil also has decreased
with the increase in curing time compared to that of sample untreated with terrazyme. It
is due to the decrease in voids after enzyme action thereby not allowing the water to flow
through the soil (Gupta etal., 2016). Terrazyme decreased liquid limit by as much as 28%
and shrinkage limit by 30% in two weeks on high liquid limit clays, but had little effect on
plastic limit (Eujine et al., 2014). Effect of terrazyme for different types of soil varies and
it mainly depends on dosage. Studies have been reporting the variations on the geotechni-
cal properties for different dosages of terraazyme for (type of soil) (Saini and Vaishnava,
2015).
231

4 BIOCEMENTATION
An alternative to the addition of chemicals such as bagasse ash and various bioenzymes
is by the use of some biomediated stabilization technique. Microbial Induced Calcite Pre-
cipitation (MICP) is one such technique wherein some bacteria are used to catalyse chemical
reactions, resulting in precipitation of calcite within the soil pores. MICP can be induced in
soils by many methods, but are usually simulated by one of the four methods—urea hydroly-
sis, denitrification, iron reduction, and sulphate reduction. Predominance of these various
mechanisms depends on their associated reaction’s propensity to occur in the environment.
Ureolysis is predominant in manipulated soil amongst others because the reaction changes
the environmental conditions of a system (i.e. increase in pH), which inhibits other competi-
tive processes. The basic concept of urea hydrolysis involves the hydrolysis of urea to produce
carbonate, which combine with the supplied calcium substrate to precipitate calcite. A bacte-
rium with an active urease enzyme is chosen for this purpose, which when introduced along
with urea, hydrolyses it into ammonia and carbonate.
CO(NH2 )2 + 2H2O → 2 NH+4 + CO32− (1)
The produced carbonate ions precipitate in the presence of calcium ions as calcite crystals,
which form cementing bridges between the existing soil grains.
Ca 2+ + CO 32− → CaCO3( s ) ↓ (2)
Calcium carbonate thus formed is precipitated at particle-particle interfaces and thus

bring improvement in engineering properties of the treated soil into picture. Further, once
precipitated, the calcium carbonate will only dissolve very slowly (at a geological time scale),
either when continuously flushed by buffered acidic groundwater or as a result of acidifying
processes in the pores (e.g. degradation of biomass). When sufficient calcium carbonate is
precipitated, durable soil stabilization can be achieved. (Van Paassen et al., 2009).
4.1 Effect of MICP on marine clays

Very few literature is available on the effect of MICP on clays, much less in marine clays.
Rebata-Landa (2007) showed a relation between grain size and CaCO3 content, and maxi-
mum carbonate deposition observed on grains was approximately 100 µm in size. This
precisely puts Kuttanad clay at a potentially viable range, as Kuttanad clay although clay
contains large quantities of silt. Vinod et al. (2017) have identified a suitable laboratory
application procedure for MICP in clays. Kannan et al. (2017) have identified optimums for
reagent concentrations and treatment durations in a laboratory scale for marine clays on the
basis of variation of liquid limit.
Tests were conducted to identify the variation of liquid limit upon treatment in Kuttanad
clays. Liquid limit tests were conducted as per IS 2720 part V – 1985 on 4 clay samples
collected from across Kuttanad region both before treatment and 15 days after treatment.
Results are tabulated in Table 3.
The clear reduction in liquid limit brought about by the treatment is indication of the
effectiveness of biomediated techniques, in particular MICP, as stabilization techniques for
Kuttanad clays.
Table 3. Effect of MICP on liquid limit of Kuttanad clay.
Sample Initial wL (%) Final wL (%)
1 118.8 88.3
2 126 95.9
3 106.4 80
4 174.2 132.3
232

5 BIOENCAPSULATION
The dredged or excavated marine clays can be strengthened by an innovative method known
as bioencapsulation, which can convert the dredged clay wastes into value added construc-
tion materials. It is a process to increase strength of soft clayey soil through the formation of
strong shell around a piece of soft material by the action of urease producing bacteria (UPB).
Thus, bioencapsulation can be a more effective alternative to MICP for improving Kuttanad
clay for applications as a fill material, especially for pavement purposes.
5.1 Mechanism of bioencapsulation

As reported in the literature of Ivanov et al., (2015), marine clay samples were subjected to
various tests such as unconfined compression tests, slaking tests etc. before and after treat-
ment. Primarily, the clay was mixed with dry biocement and then fed into a granulator to
produce 5 mm clay balls. The clay balls were immersed into a solution of calcium salt and
urea for bioencapsulation. The clay aggregates were then removed from the solution and
incubated for one day before testing. The calcium concentrations were reportedly measured
by optical emission spectrometry.
5.2 Effect of bioencapsulation on clays

Reported study shows that the unconfined compressive strength increased from zero to a con-
siderable value of about 40 kPa, which resembles the strength of a sandstone. The strength of
aggregates mainly came from the shell, as the change in strength of nucleus is not much. The
water content of clay aggregates decreased from 55% to 41% after bioencapsulation, which
was partially due to the use of water for urea hydrolysis. This decrease in the water content
led to an increase in the compressive strength. From the slaking test, it was reported that the
clay aggregates with UPB showed much higher resistance to disintegration than untreated
clay balls. But the strength of bioencapsulated aggregates decreased with an increase in the
size more than 5 mm (Ivanov et al., 2015).
6 CONCLUSIONS
This paper reviews some of the environmental friendly methods that can be potentially used
to stabilise Kuttanad clays. Consideration of soil as a living ecosystem offers the potential
for innovative and sustainable solutions to geotechnical problems. This is a new paradigm
for many in geotechnical engineering. Realising the potential of this paradigm requires a
multidisciplinary approach that embraces biology and geochemistry to develop techniques
for beneficial ground modification. The suggested biological methods possess the potential
as mentioned, but needs fine tuning to potentially apply it in a field scale. The laboratory
study showed a significant reduction in liquid limit, which underlines the potential of the said
techniques. The implementation of the same will thus take time. But when compared with the
presently used cement-based techniques, thought to be harmless in spite of the energy indu-
cive, carbon producing manufacturing process, these techniques go a long way into building
a sustainable and energy efficient future. Bioencapsulation is a modification further to the
biological method which could be an even better alternative. There are undoubtedly many
such processes yet to be discovered and further research is required to delineate them. Clear
however, that the biological processes influence engineering soil properties and is the future
of ground improvement, especially improvement of Kuttanad Clay.
REFERENCES
Eujine, G.N., Somervell, L.T., Chandrakaran, S., & Sankar, N. 2014. Enzyme Stabilization of High
Liquid Limit Clay. European Journal of Geotechnical Engineering 19: 6989–6995.
233

Gandhi, K.S. 2012. Expansive Soil Stabilization Using Bagasse Ash, International Journal of Engineer-
ing Research and Technology 1(5): 1–3.
Ivanov, V., Chu, J., Stabnikov, V., & Li, B. 2015. Strengthening of soft marine clay using biocementation.
Marine Georesources and Geotechnology 33(4): 320–324.
Kanchan, L.S., & Jawaid, S.M.A. 2013. Geotechnical Properties of Soil Stabilized with Bagasse Ash,
International Journal of Biological Research and Technological Research 1(9): 5–8.
Kannan, K., Bindu, J. & Sajna, S. 2017. Stabilisation of Marine Clays by MICP: Optimum Reagent
Concentration and Treatment Duration, Proceedings of 3rd International Conference on Materials
Mechanics and Management (IMMM 2017), Thiruvananthapuram.
Kharade, A.S., Salunkhe, S.K., & Dadage, M.D. 2014. Effective Utilization of Sugar Industry Waste
Bagasse Ash in Improving Properties of Black Cotton Soil. International Journal of Engineering
Research and Technology 3(4): 265–270.
Rebata-Landa, V. 2007. Microbial activity in sediments: Effects on soil behavior, Ph.D. dissertation,
Georgia Institute of Technology.
Saini, V. & Vaishnava, P. 2015. Soil Stabilization by Using Terrazyme. International Journal of Advances
in Engineering & Technology 8(4): 566–573.
Shirsath, H.A., Joshi, S.R., & Sharma, V. 2017. Effect of Bio-Enzyme (Terrazyme) on the Properties of
Subgrade Soil of Road. Proceedings of the International Conference on Recent Trends in Civil Engi-
neering, Science and Management: 53–58.
Van Paassen, L.A., Harkes, M.P., van Zwieten, G.A., Van der Zon, W.H., Van der Star, W.R.L., &
Van Loosdrecht, M.C.M. 2009. Scale up of BioGrout: a biological ground reinforcement method.
Proceedings of the 17th International Conference on Soil Mechanics and Geotechnical Engineering:
2328–2333.
Vinod, P., & Bindu, J. 2010. Compression Index of Highly Plastic Clay—an Emperical Correlation.
Indian Geotechnical Journal 40(3): 174–180.
Vinod, P., Bindu, J. and Kannan, K. 2017. Potential Utilisation of MICP in clays; Development of a
Suitable Application Method, Proceedings of International Conference on Geotechniques for Infra-
structure Projects (GIP 2017), Thiruvananthapuram.
234

Development of critical state line concept from hypoplastic model

simulations on triaxial strength of silty sand
M. Akhila, Kodi Ranga Swamy & N. Sankar

Department of Civil Engineering, NIT Calicut, Kerala, India
ABSTRACT: Earthquakes can cause liquefaction-induced damages of civil infrastructures

supported on the foundation soil. Hence it is required to analyze the susceptibility of lique-
faction in foundation soils including sands, silts, and low plastic clays. Based on Critical State
Line (CSL) concept, it is possible to identify whether the soil can be susceptible to liquefac-
tion or not. The critical state line can be developed based on drained triaxial compression
testing data. In this study, CSL of silty sand is developed from drained triaxial test simula-
tions based on the hypoplastic model.
1 INTRODUCTION
The liquefaction may occur in fully saturated sands, silts, and low plastic clays. When the
saturated soil mass is subjected to seismic or dynamic loads, there is a sudden build-up of
pore water within a short duration. If the soil could not dissipate the excess pore pressure,
it will result in a reduction of the effective shear strength of soil mass. In this state, the soil
mass behaves like a liquid and causes large deformations, settlements, flow failures, etc. This
phenomenon is called soil liquefaction. As a result, the ability of soil deposit to support the
foundations of buildings, bridges, dams, etc. are reduced. Liquefiable soil also exerts a higher
pressure on retaining walls, which can cause them to tilt or slide. The lateral movement could
prompt settlement of the retained soil and distraction of structures constructed on various
soil deposits. A sudden build-up of pore water pressure during earthquake also triggers land-
slides and cause the collapse of dams. Liquefaction effects on damages of structures are com-
monly observed in low-lying areas near the water bodies such as rivers, lakes, and oceans.
The CSL is the boundary line which separates the liquefiable and non-liquefiable soil
states (Kramer, 1996). Vu To-Anh Phan et al. (2016) have studied the critical state line (CSL)
of sand-fine mixtures experimentally. Their result indicates that there is a unique line CSL
obtained with specific fines content for various confining pressures and different initial glo-
bal void ratios. Aminaton Marto et al. (2014) suggested that neither the fines percentages
nor other corresponding compositional characteristics are adequate to be correlated with
the critical state parameters of sand matrix soils. Sadrekarimi A and Olson S M (2009) used
ring shear tests to find the CSL of soils as the limited displacement that the triaxial device
is capable of imposing on a specimen is insufficient to reach a critical state where particle
rearrangement and potential crushing are complete. The present paper attempts to use CSL
developed from the hypoplastic model simulations to analyze the liquefaction susceptibility
of silty sand under static triaxial loading.
2 HYPOPLASTIC MODEL
It is found that hypoplastic models are more advanced to elastic-plastic models for con-
tinuum modelling of granular materials. In contrast to elastoplastic models, a decomposition
of deformation components into elastic and plastic parts, yield surface, plastic potential,
235

flow rule and hardening rule are not needed in the hypoplastic model. The hypoplastic con-
stitutive law represents the deformation behaviour of cohesionless soils up to certain fines,
including the nonlinearity and inelasticity. The first version of the hypoplastic constitutive
model was proposed by Kolymbas (1985). It used a single state variable, the current Cauchy
stress Ts. Later another state variable the void ratio e was added. The hypoplastic constitutive
equation in general form is given by:
Ts = F (Ts , e, D ) (1)

Herein Ts represents the objective stress rate tensor as a function of the current void ratio
e, the Cauchy granulate stress tensor Ts and the stretching tensor of the granular skeleton D.
D = (L + LT)/2 is the symmetric part of the deformation gradient L = ∂v (xx ,t ) , v being the veloc-
ity vector of the continuum representing the grain skeleton at a point x.
3 MODEL PARAMETERS
The silty sand used in this study is processed by mixing 40% quarry dust into the fine sand.
The fine sand is collected from Cherthala, Kerala and quarry dust is procured from Blue
Diamond M-sand manufacturers, Kattangal, Kerala. All the basic properties tests were
performed on the soil combinations, and the properties are listed in Table 1. A combined
dry sieve and hydrometer analysis were carried out to obtain the particle size distribution
(Figure 1).
The standard routine laboratory tests were conducted on the non-plastic silty sand to
determine the model parameters. The limited eight numbers of hypoplastic model para
meters (ϕc, hs, n, ei0, ed0, ec0, β, and α) are determined based on the detailed procedure for the
determination of model parameters explained by Herle and Gudehus (1999).
3.1 Critical state friction angle

The critical friction angle ϕc can be obtained from the angle of repose of dry granular mate-
rial from cone pluviation tests. If the portion of grains with diameter <0.1 mm is too large,
conventional shear tests are recommended for determination of ϕc. Hence, for the present
study, direct shear box tests were conducted on loose soil at different consolidation pressures
to determine the critical friction angle ϕc. Direct shear box tests were conducted on silty sand
Table 1. Basic properties of silty sand.
Index property Value
Specific gravity 2.69

D10 (mm) 0.03
D50 (mm) 0.15
Cu 7.33
Cc 0.74
γmax (kN/m3) 17.56
γmin (kN/m3) 13.93
Figure 1. Particle size distribution curve of silty sand.
236

samples of loosest and densest possible conditions at different consolidation pressures. The
critical friction angle of a sample is determined from the slope of linear failure envelope in
loosest possible state.
3.2 Minimum, maximum and critical void ratios at zero pressure state
Based on limited densities, the maximum and minimum void ratios were estimated by using
the empirical equations. Three limit void ratios at zero pressures, i.e., ei0 (during isotropic
compression at the minimum density), ec0 (critical void ratio) and ed0 (maximum density) were
estimated.
3.3 Peak state coefficient

The exponential coefficient of peak state was determined after conducting a series of direct
shear box tests on the silty sand in densest possible condition at different consolidation pres-
sures. The peak friction angles of each sample are determined from the slope of linear failure
envelope in the densest possible state, and they used to estimate the peak state exponent coef-
ficient of model parameter (α).
3.4 Hardness coefficient and exponent “n.”

Oedometer compression tests were carried out on the silty sand in loosest possible condition
to arrive the hardness parameters. From this experimental e-log p curve and by using math-
ematical formulations the hardness parameters of silty sand was determined.
3.5 Stiffness parameter

Oedometric compression tests were carried out on the silty sand in loosest and densest pos-
sible conditions (see Figure 2) to arrive the stiffness parameter. From the experimental e-log
σ curves and by using mathematical formulations the stiffness parameter is found. All the
model parameters are listed in Table 2.
Figure 2. Particle experimental e-log p curves of silty sand in both loosest and densest states.
Table 2. Model parameters.
Parameter Value
φc 33°
edo 0.413
eco 0.890
eio 1.068
Α 0.035
hs 43 MPa
N 0.509
Β 0.5
237

4 VALIDATION OF THE MODEL
The element test program has been prepared by Herle using mathematical formulations
involved in Hypoplastic soil model. The test program requires three input files namely mate-
rial parameters, initial state parameters and test conditions. Initially, the hypoplastic model
simulations were performed on oedometric compression of both the loose and dense silt soil
samples.
The overlapped curves of both the experimental as well as model simulated tests are pre-
sented to check the validity of the model. Figures 3 shows the combined overlapped e-log
p curves under oedometeric loading. It can be seen that the model simulation results well
coincide with experimental curves.
Before performing the numerical simulations on consolidated triaxial compression, the
numerical model is validated with the experimental data of CD triaxial test conducted on the
silty sand at a void ratio of 0.5 corresponding to the relative density of 85%. The overlapped
stress-strain relationships on silty sand from both the experimental and numerical model
are presented in Figure 4. It can be seen that the model simulation results well coincide with
experimental curves. Therefore, the present model study is extended to perform the triaxial
loading simulations under the drained conditions to examine the liquefaction susceptibility
of silty sand based on CSL concept.
Figure 3. e-log p curves on loose and dense silty sand under oedometeric compression loading.
Figure 4. Comparison of experimental and numerically simulated stress-strain relationship on the silty
sand at e = 0.5.
238

5.1 Effect of density

To study the effect of void ratio on the drained response of silty sands, a series of static tri-
axial test simulations were performed on silty sand subjected to consolidation pressures of
100, 200 and 400 kPa by varying the void ratios in the range of 0.5 to 0.98. (Due to space
limitations, only results of 200 kPa are presented as graphs in this paper). This range of
void ratios was chosen in such a way that the soil state has to change from very loose to the
dense condition. It is observed that the dense silty sands experience the dilation behaviour
by indicating the sharp peak deviator stress and increase in the volume of soil, i.e., undergo
a sudden expansion.
However, loose silty sands experience the contraction behaviour that shows the continuous
decrease in volume, i.e., compression. At all tested consolidation pressures, the soil behaviour
changes from dilation to contraction (liquefaction) state with an increase in the void ratio.
In dense silty sands, it exhibits initial contraction up to a certain axial strain limit and then
causes to volumetric expansion leads to dilation. The reason may be due to the re-adjustment
of solid particles.
The deviator stresses are increasing with decrease in void ratios at the same axial strain
level. A sharp peak stress was observed for dense silty sands at the low strain levels of 7–12%
and strain level at the peak stress is increasing with increase the consolidation pressure. For
loose silty sands (ec = 0.98∼0.804), the constant ultimate stress is reached at the limited failure
strain of 20% level. Herein, 20% failure strain is assumed as the critical state level that indi-
cates where the deviator stress almost becomes constant. For the medium dense silty sands,
(ec = 0.65), a slight peak deviator stress is observed at low strain levels and further decreased
to residual values.
From the volume change responses of silty sand at different void ratios consolidated under
each pressure application of 100, 200 and 400 kPa respectively. It demonstrates that the loose
silty sands are exhibiting fully contraction behaviour, i.e., volume reduction. The dilative
behaviour, i.e., volume expansion is increasing with decrease in the void ratios from 0.65 to
0.5 representing the state of soil changes from medium dense to very dense state. In loose
silty sands, the quantitative values of shear strength and volume changes are increasing with
increasing the applied consolidation pressures. However, in medium and dense silty sands,
Figure 5. (a) Stress-strain characteristics and (b) volume change response of silty sand (at different
void ratios and σ3 = 200 kPa).
239

the amount of expansion volume change is decreases with increasing the applied pressures.
In summary, the result concludes that the drained response, i.e., either contraction or dilation
(liquefaction susceptibility) is dependent on density/void ratio of the soils consolidated at
particular low, medium, and high pressures.
5.2 Effect of consolidation pressure

To study the effect of consolidation pressure on the drained triaxial response of loose, and
dense silty sands, the numerical simulations were performed on silty sands consolidated at
different pressures in the range of 50 to 400 kPa. The response of loose silty sand is presented
in Figure 6 and discussed here.
Figure 6 shows the drained response in loose (ec = 0.98) silty sands at different consolida-
tion pressures. Figure 6(a) shows the stress-strain characteristics of silty sand consolidated at
different pressures. It can be inferred from the figure that the deviator stresses are increases
with increase in applied consolidation pressures. It is observed that the ultimate residual con-
stant stress state was reached at about 20% strain level. In the previous section, the effect of
density on the drained response is explained in terms of stresses because the tested samples
are subjected to unique confining pressures. The effect of density on trends of stresses (stress
ratios) is similar due to the unique applied pressure. However, the effect of confining pressure
is to be expressed in terms of normalized stress ratios due to the test samples are subjected
to different pressures.
For example, the effect of confining pressure on the drained response in terms of stress-
strain relation shown in Figure 6(a) is compared with drained response in terms of normal-
ized stress ratios varies with axial strain shown in Figure 6(b). It can be seen from Figure 6(b)
Figure 6. (a) Stress-strain characteristics, (b) Stress ratio characteristics and (c) Volume change
response (loose silty sand consolidated at different pressures).
240

Figure 7. Identification of liquefaction susceptibility from CSL.
that the normalized stress ratios are decreasing with an increase in consolidation pressures
indicate the reverse trends compared to Figure 6(a). In loose silty sands, the ultimate con-
stant deviator stress ratios are obtained at a large strain level of 20% that indicates the silty
sand behaves as contraction and stress ratio is more at low consolidation pressures.
Figure 6(c) presents the volume change response of silty sand consolidated at different
pressures. It demonstrates that the low consolidated silty sands (σc = 50 kPa) exhibiting less
contraction behaviour i.e., less volume reduction. The stress ratios are high at low consolida-
tion pressures due to less contraction behaviour. However, the more contraction behaviour
is observed at a high consolidated pressure of 400 kPa. It indicates the volume reduction is
increasing with increase in applied pressures from 50 to 400 kPa. The result concludes that
the soil behaviour changes from less contraction to highly contraction state in loose silty
sands while increasing the applied pressures. The high contraction may take place due to the
crushing of particles under the application of high pressures. The contraction soils are not
stable and susceptible to liquefaction.
5.3 Critical state line

From the hypoplastic simulations, critical state void ratios are determined at constant volume
change corresponding to 20% failure strain level under each application of consolidation
pressure on the silty sand in loosest possible condition (ec = 0.98). CSL is the line connect-
ing the critical state void ratios at consolidation pressures in the range of 50 to 4200 kPa. As
the range of the consolidation pressures is very large, for better representation, a log scale is
required to make the linear relation between pressures and void ratios as shown in Figure 7.
It can be seen that the initial critical void ratio at loosest state decreases with increasing
the applied pressures consolidated by the soil. The trends are well matched with published
research works (Kramer, 1996).
In this study, CSL of silty sand is developed from drained triaxial test simulations based on
the hypoplastic model. The major findings from the study are given below:
• The effect of density on the drained response of contraction and dilation state of silty sand
depends on the applied range of consolidation pressures; similarly, the effect of consolida-
tion pressure on the response of contraction and dilation state of silty sand again depends
on denseness of soil.
• The dense silty sands experience the dilation behaviour by indicating a continuous increase
in deviator stress to higher values and decrease in pore water pressure towards negative
values. However loose silty sands experience the contraction behaviour that shows the
continuous increase in pore water pressures and reduction in deviator stress.
• A sharp peak stress was observed for loose silty sands at the low strain levels of 2–4% and
then further decreases towards residual stress levels. For dense and medium dense silty sands
the continuous increase in deviator stress was observed up to the failure strain level of 25%.
241

• The CSL was developed from the triaxial drained response of silty sands. CSL act as the
boundary to separate the state of no liquefaction and liquefaction regions. The soil behav-
iour change from the dilative state (No liquefaction) to contraction state (Susceptibility of
liquefaction) depends on both the void ratio and applied consolidation pressure.
REFERENCES
Aminaton Marto, Choy Soon Tan, Ahmad Mahir Makhtar and Tiong Kung Leong. 2014. Critical State
of Sand Matrix Soils. The Scientific World Journal, Hindawi Publishing Corporation.
Atkinson, John, H., and Bransby, P.L. 1978. The Mechanics of Soils: An Introduction to Critical State
Soil Mechanics, McGraw-Hill.
Gudehus, G. 1996. A comprehensive constitutive equation for granular materials. Soils and Foundations,
36(1), 1–12.
Herle, I., and G. Gudehus. 1999. Determination of parameters of a hypoplastic constitutive model from
properties of grain assemblies. Mechanics of Cohesive-Frictional Materials, 4, 461–486.
Kolymbas, D. 1985. A generalized hypoelastic constitutive law. In Proceedings of the eleventh Interna-
tional Conference on Soil Mechanics and Foundation Engineering.
Sadrekarimi, A., and Olson, S.M. 2009. Defining the critical state line from triaxial compression and
ring shear tests. In Proceedings of the 17th International Conference on Soil Mechanics and Geotechni-
cal Engineering: The Academia and Practice of Geotechnical Engineering, 1, 36–39.
Steven, L., Kramer. 1996. Geotechnical Earthquake Engineering. Prentice Hall, New Jercy.
Vu To-Anh Phana, Darn-Horng Hsiaob and Phuong ThucLan Nguyenc. 2016. Critical State Line and
State Parameter of Sand-Fines Mixtures, Sustainable Development of Civil, Procedia Engineering—
Urban and Transportation Engineering Conference, 142, 299–306.
242

Design of an embankment laterally supported with secant piles

at Kuttanad
J. Jayamohan & N.R. Arun

LBS Institute of Technology for Women, Thiruvananthapuram, India
K. Balan
Rajadhani Institute of Engineering and Technology, Thiruvananthapuram, India
S. Aswathy Nair, L.K. Vaishnavi, Megha S. Thampi, D.R. Renju & Chithra Lekshmi
ABSTRACT: This paper presents a case study of the investigation of recurring breaches of
the embankment at Puthenarayiram Padasekharam, Kuttanad and the design of its reconstruc-
tion. At a location called Kundarikund in D Block of Kuttanad, a section of embankment
frequently collapses, inundating the cultivable land. A hydrographic survey was conducted to
determine the bed profile and the velocity of water was measured using a current meter. The
presence of two depressions on the paddy field side of the embankment where the breach occurs
is detected. Laboratory tests on soil samples obtained from boreholes indicated a very high void
ratio and the water content was much higher than the liquid limit. Hence, the insitu undrained
shear strength of the soil was determined by conducting a field vane shear test. The global and
internal stability were checked as per standard geotechnical practices. Only very less passive
resistance could be generated in the weak clay. To enhance the passive resistance, a berm made of
Geobags filled with soil was provided on the paddy field side of the embankment. Contiguous/
secant piles were provided on the right face of the embankment throughout the breached por-
tion. At the canal side, soldier piles are provided at a spacing of two meters connected together
by RCC (Reinforced Cement Concrete) precast slabs. The secant piles on the right side are con-
nected to the soldier piles on the left side by transverse RCC beams at a spacing of two meters.
Keywords: Kuttanad, secant piles, field vane shear test, soldier piles, Geobags, berm
1 INTRODUCTION
Kuttanad is a region covering the Alappuzha and Kottayam Districts, in the state of Kerala,
well known for its vast paddy fields and geographical peculiarities. The region has the lowest
altitude in India, and is one of the few places in the world where farming is carried around 1.2
to 3.0 meters below sea level. Kuttanad is historically important and is a major rice producer
in the state. Farmers of Kuttanad are famous for bio saline farming. Kuttanad clay is a soft
soil with associated problems of low shear strength and compressibility (Vinayachandran
et al., 2013). The soil has a unique combination of minerals such as metahalloysite, kaolinite,
iron oxides and aluminum oxides. The diatom frustules present in the soil indicate biological
activity during the sediment formation and this also accounts for the nature of organic mat-
ter predominantly present in the soil, which is mostly derived from planktonic organisms.
A considerable amount of organic matter is present in the soil and the magnitude measured
accounts for about 14% by mass (Suganya & Sivapullaiah, 2015, 2017). Kuttanad soil is
expansive clay having a high void ratio and low density.
The cultivable land of Kuttanad is 1.2 to 3 meters below mean sea level. These lands are
kept submerged for about six to eight months of the year; during which a lot of organic
243

matter settles on the soil making it fertile. During the farming season, the water in the paddy
fields is pumped out into the canal which joins a lake. The water level in the canal will be
much higher than the adjoining paddy fields. The paddy fields and canal are separated by
an embankment constructed on this weak soil. The embankments are usually constructed
by driving precast Reinforced Cement Concrete soldier piles at a spacing of two meters con-
nected to each other by RCC precast slabs. These embankments are usually four meters wide
and three meters high. The construction and maintenance of these embankments are carried
out by the Irrigation Department, Government of Kerala. At a location called Kundarikund
in D Block of Kuttanad, a section of embankment collapses quite often, inundating the
cultivable land. The government spends crores of Rupees every year on restoration works.
The Irrigation Department requested the Department of Civil Engineering, LBS Institute
of Technology for Women, to study the matter and to suggest remedial measures.
2 SITE INSPECTION
The site was inspected on 31-05-2017 by the authors along with the Engineers from Irriga-
tion Department. From the site inspection, it was observed that the Pile-Slab system and the
coconut piles are tilted towards the paddy area. Water was flowing from the paddy area to
Kochar river. The surface of the water body in the paddy area is very calm and hence possi-
bility of under currents cannot be ruled out. From the observation it is presumed that surface
of soil profile in the paddy area is much deeper than that shown in the drawings provided.
One of the possible solutions which can be practically implemented cost effectively is the
pile-slab system supported by inclined piles. The inside portion of the pile-slab system should
be filled with clay reinforced with woven coir geotextiles (700 gsm).
To assess whether this solution is practically possible or not, the depth of soil in the paddy
area from water surface needs to be determined; for which a hydrographic survey was carried
out.
It is ideal to also know the water velocity as turbulent flow may occur at the site. In order
to design the inclined pile, the undrained shear strength of the soil is also required. This was
obtained by conducting a field vane shear test on the natural soil below the existing bund.
3 HYDROGRAPHIC SURVEY
A hydrographic survey was conducted to determine the bed profile and the velocity of water was
measured using a current meter. The contour map of the bed obtained from the hydrographic
survey is presented in Figure 1. The results of the hydrographic survey revealed the presence of
Figure 1. Contour map of bed surface obtained from the hydrographic survey.
244

two depressions on the right side of the embankment where breaches frequently occurs. The
maximum depth of the depression was up to five meters. The velocity of water flow was meas-
ured with a current meter. The maximum velocity of flow was observed to be 0.824 m/sec.
4 SOIL INVESTIGATION
4.1 Soil properties

The results of laboratory tests on soil samples are presented in Table 1.
4.2 Field vane shear test

Laboratory tests on the soil samples obtained from the boreholes indicated a very high void
ratio (2.36) and the water content was much higher than the liquid limit. Hence, the insitu
undrained shear strength of the soil was determined by conducting a field vane shear test
as shown in Figure 2. A bore hole of 30 cm diameter was drilled up to the base level of the
embankment using a helical auger.
The blade of the vane shear apparatus had a width of 50 mm and depth of 100 mm. The
vane was inserted at the bottom of the borehole and the test was carried out as per IS 4434 -
1997. The undrained shear strength obtained was 9 kPa, which is a very low strength.
5 DESIGN OF EMBANKMENT
Based on the observations from the hydrographic survey, field vane shear test and the soil
properties, the following recommendations are made.
Table 1. Soil properties.
Properties Clay
Class MH-OH
Natural moisture content (%) 116
Bulk density (g/cc) 1.35
Void ratio 2.36
Porosity (%) 70
Liquid limit (%) 71
Plastic limit (%) 38
Plasticity index (%) 32
Cohesion (kPa) 0.88
% Clay & silt 94.3
% Sand 5.6
Figure 2. Field vane shear test.
245

5.1 Reconstruction of the collapsed portion of the bund (length 50 m)
River side
Provide RCC precast soldier piles of cross section 35 × 30 cm and length of 10 m at a spac-
ing of 2 m, connected by RCC precast slabs. The top of the soldier piles may be connected
together by an RCC beam with a cross section of 40 × 40 cm.
A small berm (Pilla Bund) with a top width of 2 m may be constructed by driving a row of
coconut piles and filling with soil. At the external side of the Pilla Bund, woven multifilament
polypropylene geotextile must be provided at the external surface as shown in the drawing.
The specifications of this geotextile are as follows:
Tensile strength—Machine direction – 55 kN/m, Cross machine direction – 40 kN/m,
Trapezoidal tear strength—Machine direction – 0.73 kN, Cross machine direction –
0.52 kN,
Puncture Strength – 0.62 kN, Mass per unit area – 240 g/m2,
Average opening size – 0.15 mm.
Padasekharam side
The major stabilizing force is the passive pressure from the soil on the Padasekharam side. To
utilize the passive resistance of the soil, there must be a continuous wall underneath the bed
surface for a suitable depth. Hence, it is recommended to install driven precast RCC contigu-
ous piles with a cross section of 40 cm × 40 cm and length of 12 m. These contiguous piles
must be installed for a length of 50 m including the breached portion of the bund. The top of
the contiguous piles must be joined together by providing an RCC beam of 60 × 40 cm. The
beam must be connected to the RCC top beam at the river side by transverse beams at 2 m
intervals. To increase the passive resistance from the paddy side, a berm, with geobags filled
with locally available clayey soil, must be constructed.
RCC precast contiguous/secant piles

The RCC precast contiguous piles of length 12 m may be driven in two stages. At first, a pre-
cast pile of length 8 m may be driven to the correct alignment, with the top 1.2 m projecting
above water level. Then, the top 1 m portion must be mechanically chipped off. Care must be
taken that the reinforcement in the pile does not get damaged due to this chipping of concrete.
In the second stage, a 4 m pile should be cast with 1 m development length into the first stage
(total length of cast in situ portion is 5 m including the development length portion). The
reinforcements of the first stage and second stage must be welded together throughout the lap
length (1 m) with a 6 mm weld so as to have a better resistance to the impact which occurs dur-
ing driving. The finished pile should be driven only after 28 days from the concreting of the
second stage. The concrete of second stage must be cured as per relevant IS codes of practice.
Geobags
The geobags must have a width of 1.2 m, length of 1.2 m and height of 1 m. There should
be four lifting points with two straps on each lifting point. The tensile strength of each strap
should at least be 15 kN. The seam strength should be 25 kN/m. The fabric of the geobag
should be woven polypropylene with a density of 325 gsm. The fabric should have a wide
width tensile strength of 55 kN/m in both machine and cross machine directions. The CBR
puncture strength should be 6 kN.
Calculations
The in situ undrained shear strength of the soil is very low. If the passive resistance of clay
alone is considered, the length of the piles will be very large and will not be feasible. Hence, it
is required to enhance the passive resistance by providing a berm. The thickness of the berm
proposed is 3 m. The top width of the berm was determined based on the width required for
the formation of a full passive wedge. The depth of embedment ‘D’ below the bottom level of
berm is calculated by taking moments of all the forces about top. The details of the various
forces acting on the wall are shown in the pressure diagram (Figure 3).
246

Figure 3. Pressure diagram.
φ = 1°, c = 8.83 kPa
ka = 0.966
kp = 1.04
Unit weight of soil, γ = 13 kN/m3
Saturated unit weight, γsat = 14.3 kN/m3
Submerged unit weight, γ’ = 4.5 kN/m3
Taking moments about top of the pile and simplifying we get
0.75D3 + 11.06D2 − 16.04D − 286.12 = 0
Solving, D = 5 m
The total depth works out to be 10.75 m. However a total length of 12 m from the top of
the bund may be provided for the pile including factor of safety.
The top width of the berm was determined based on the width required for the formation
of a full passive wedge. The minimum top width comes to 5.20 m. However the top width
provided is 7 m, considering the factor of safety. The bottom width was determined based on
the angle of repose from the top outer edge.
The time schedule to be followed for filling of soil inside the bund is
Stage-1: First one meter filling inside the bund
Stage-2: Second one meter after two months from the completion of first layer
Stage-3: Third one meter after two months from the completion of second layer
Stage-4: Top (fourth) layer after six months from the completion of third layer
The berm must also be constructed simultaneously during the filling of the bottom layers
inside the bund.
The soil for filling the bund/berm must be taken from sites at least 150 m away from the
bund.
The soil fill inside the bund must be reinforced with woven coir geotextile of 700 gsm with
12.5 mm opening size. The vertical spacing of reinforcement must be 1 m. The cross section
of the embankment is presented in Figure 4.
5.2 R
econstruction of the bund for a length of 35 m on either sides of collapsed portion
(length 70 m)
Provide RCC precast soldier piles with a cross section of 30 × 30 cm and length 7.8 m at a
spacing of 2 m, connected by RCC precast slabs. The top of the soldier piles must be con-
nected together by an RCC beam with a cross section of 40 × 40 cm.
Berms (Pilla Bund) with a top width of 2 m (similar to the collapsed portion) must be
constructed by driving a row of coconut piles and filling soil on both sides of the bund as
detailed in Figure 5.
Woven multifilament polypropylene geotextile must be provided at the external surface as
shown in Figure 4. The specifications of this geotextile are as follows:
247

Figure 4. Cross section of proposed embankment at the breached portion.
Figure 5. Cross section of proposed embankment on either sides of breached portion for a length of
35 m.
Tensile strength—Machine direction – 55 kN/m, Cross machine direction – 40 kN/m,

Trapezoidal tear strength—Machine direction – 0.73 kN, Cross machine direction –
0.52 kN,
Puncture Strength – 0.62 kN, Mass per unit area – 240 g/m2,
Average opening size – 0.15 mm.
6 CONCLUSIONS
The reasons for the recurring breaches of the embankment at Puthenarayiram Padasekharam,
Kuttanad have been investigated. The findings are:
• The presence of two depressions of about 5 m in depth on the paddy field side of the
embankment were hampering the global stability
• The pile-slab construction method practiced by the Irrigation Department could not har-
ness the passive resistance of the soil below the embankment
• A revised design of embankment has been proposed with secant piles and a berm on the
paddy field side.
REFERENCES
Suganya, K. & Sivapullaiah. P.V. Effect of changing water content on the properties of Kuttanad soil,
Geotechnical and Geological Engineering 33, 913–921.
Suganya, K., & Sivapullaiah. P.V. Role of composition and fabric of Kuttanad clay: a geotechnical per-
spective, Bulletin of Engineering Geology and the Environment 76, 371–381.
Vinayachandran. N., Narayana. A.C., Najeeb. K.M., & Narendra. P. (2013). Disposition of aquifer
system, geo-electric characteristics and gamma-log anomaly in the Kuttanad alluvium of Kerala,
Journal of the Geological Society of India 81, 183–191.
248

Effect of anchorage of geosynthetic reinforcement on the

behaviour of reinforced foundation bed
Aleena Mariam Saji, Alen Ann Thomas, Greema Sunny, J. Jayamohan & V.R. Suresh
ABSTRACT: Geosynthetics demonstrate their beneficial effects only after considerable settle-
ment, since the strains occurring during initial settlements are insufficient to mobilize significant
tensile load in the geosynthetic. This is not a desirable feature since for foundations of certain
structures, the values of permissible settlements are low. Anchoring the reinforcement is a prom-
ising technique yet to be comprehensively studied. This paper presents the results of a series of
finite element analyses carried out to investigate the improvement in load-settlement behaviour
of a strip footing resting on a Reinforced Foundation Bed due to anchoring the Geosynthetic
Reinforcement. It is observed that the bearing capacity can be considerably increased without the
occurrence of excessive settlement by anchoring the geosynthetic reinforcement with micropiles.
Keywords: Geosynthetics, Finite Element Analyses, Anchoring, Load-Settlement Behav-

iour, Reinforced Foundation Bed
1 INTRODUCTION
The decreasing availability of proper construction sites has led to the increased use of mar-
ginal ones, where the bearing capacity of the underlying deposits is very low. By the applica-
tion of geosynthetics it is possible to use shallow foundations even in marginal soils instead
of expensive deep foundations. This is done by either reinforcing cohesive soil directly or
replacing the poor soils with stronger granular fill in combination with geosynthetic rein-
forcement. In low-lying areas with poor foundation soils, the geosynthetic reinforced foun-
dation bed can be placed over the weak soil. The resulting composite ground (reinforced
foundation bed) will improve the load carrying capacity of the footing and will distribute
the stresses on a wider area on the underlying weak soils, hence reducing settlements. Dur-
ing the past 30 years, the use of reinforced soils to support shallow foundations has received
considerable attention. Many experimental and analytical studies have been performed to
investigate the behaviour of reinforced foundation beds for different soil types (eg. Binqet
and Lee (1975), Shivashankar et al. (1993)). Several experimental and analytical studies were
conducted to evaluate the bearing capacity of footings on reinforced soil (eg. Shivashankar
and Setty (2000); Shivashankar and Reddy (1998); Madhavilatha and Somwanshi (2009);
Alamshahi and Hataf (2009); Vinod et al. (2009) Arun et al. (2008) etc).
It is now known that geosynthetics demonstrate their beneficial effects only after con-
siderable settlements, since the strains occurring during initial settlements are insufficient
to mobilize significant tensile load in the geosynthetic. This is not a desirable feature since
for foundations of certain structures; the values of permissible settlements are low. Thus
there is a need for a technique which will allow the geosynthetic to increase the load bearing
capacity of soil without the occurrence of large settlements. Lovisa et al 2010 conducted
laboratory model studies and finite element analyses on a circular footing resting on sand
reinforced with prestressed geotextile. It was found that the addition of prestress to reinforce-
ment resulted in a significant improvement in the load bearing capacity and reduction in set-
tlement of foundation. Lackner et al. (2013) conducted about 60 path controlled static load
249

displacement tests and 80 cyclic load displacement tests to determine the load-displacement
behaviour of prestressed reinforced soil structures. Unnikrishnan and Aparna Sai (2014) car-
ried out extensive studies on footings in clay supported on encapsulated granular trench.
In this research, the effects of anchoring the geosynthetic reinforcement, on the load-bearing
capacity and settlement response of a reinforced foundation bed overlying weak soil are investi-
gated by carrying out nonlinear finite element analysis using the FE software PLAXIS 2D. The
effects of anchoring the reinforcement on the stress distribution at the interface between foun-
dation bed and weak soil, axial force distribution in the reinforcement, stress distribution at the
interface between reinforcement and surrounding granular soil etc. are particularly studied.
2 FINITE ELEMENT ANALYSES
Finite element analyses are carried out using the commercially available finite element soft-
ware PLAXIS 2D. For simulating the behaviour of soil, different constitutive models are
available in the FE software. In the present study Mohr-Coulomb model is used to simu-
late soil behaviour. This non-linear model is based on the basic soil parameters that can be
obtained from direct shear tests; internal friction angle and cohesion intercept. Since strip
footing is used, a plain strain model is adopted in the analysis. The settlement of the rigid
footing is simulated using non zero prescribed displacements.
The displacement of the bottom boundary is restricted in all directions, while at the verti-
cal sides; displacement is restricted only in the horizontal direction. The initial geostatic stress
Figure 1. Geometric model.
Figure 2. Typical deformed shape.
Table 1. Properties of soil.
Property Sand Clay
Dry Unit Weight (kN/m3) 17 16

Specific Gravity 2.3 2.1
Cohesion (kPa) 2 20
Angle of Shearing Resistance (°) 30 0
Modulus of Elasticity (kPa) 15000 7000
Poisson’s Ratio 0.25 0.3
250

states for the analyses are set according to the unit weight of soil. The soil is modelled using
15 noded triangular elements. Mesh generation can be done automatically. Medium mesh
size is adopted in all the simulations. The size of the strip footing (B) is taken as one metre
and the width and depth of soil mass are taken as 10B in all analyses.
The reinforcement is modelled using the 5-noded tension element. To simulate the interac-
tion between the reinforcement and surrounding soil, an interface element is provided on both
upper and lower surface of reinforcement. The interaction between soil and reinforcement is
simulated by choosing an appropriate value for strength reduction factor Rinter at the interface.
The geometric model is shown in Fig. 1 and the typical deformed shape in Fig. 2. The soil is
modeled using 15-node triangular elements. Poisson’s ratio of the soil is assumed to be 0.25 for
all cases. Properties of locally available sand and clay are adopted in the analyses. The mate-
rial properties adopted are outlined in Table 1. The properties required for the geosynthetic is
Elastic Axial Stiffness (EA) and the value adopted is 40 kN/m. Footing is modelled as a plate.
The anchorage provided by micropiles is modelled with fixed end anchors. The effect of
change in spacing and length of micropiles is simulated by varying the stiffness of fixed end
anchors. Various values of stiffness (EA) of the anchors adopted in the analyses are 1000,
1500, 2000 and 3500 MN/m respectively.
3.1 Load-settlement behaviour

Vertical Stress vs Settlement curves for strip footing resting on Reinforced Foundation Bed
with anchored geosynthetic reinforcement; obtained from FEA, for various values of stiff-
ness of anchor are presented in Figure 3. It is seen that the load-settlement behaviour consid-
erably improves due to anchoring the reinforcement.
3.2 Improvement factor

To quantify the improvement in load-settlement behaviour attained due to anchoring the
reinforcement, an Improvement Factor (If) is defined as the ratio of stress with anchored
reinforcement to that with unanchored reinforcement; at 2 mm settlement.
Figure 4 presents the variation of Improvement Factor with Stiffness of Anchor, obtained
from Finite Element Analyses. It is seen that as the stiffness of anchor increases, the improve-
ment factor increases.
When the footing settles the mid portion of the reinforcement moves down and its edges
moves inwards and takes a deformed shape as shown in Figure 2. When the reinforcement is
anchored at its edges, there will be an additional resistance to this deformation and more tensile
stress gets mobilized in it. This additional mobilized tensile stress improves the load-settlement
Figure 3. Vertical stress vs settlement curves.
251

Figure 4. Improvement factor vs stiffness of anchor.
behaviour. It is seen from Figures 3 and 4 that as the stiffness of anchor increases, the improve-
ment in load-settlement behaviour increases.
4 CONCLUSIONS
Based on the results of Finite Element Analyses carried out the following conclusions are
drawn.
• Anchoring the Geosynthetic Reinforcement considerably improves the load-settlement
behaviour
• The improvement factor increases with stiffness of the anchor
REFERENCES
Alamshahi, S. and Hataf, N. (2009). Bearing capacity of strip footings on sand slopes reinforced with
geogrid and grid-anchor, Geotextiles and Geomembranes, 27 (2009) 217–226.
Arun Kumar Bhat, K., Shivashankar, R. and Yaji, R.K. (2008), “Case study of land slide in NH 13
at Kethikal near Man galore, India”, 6th International Conference on Case histories in Geotechnical
Engineering, Arlington, VA, USA, paper no. 2.69.
Binquet, J. and Lee, K.L. (1975). Bearing capacity tests on reinforced earth slabs. Journal of Geotechni-
cal Engineering Division, ASCE 101 (12), 1241–1255.
Lackner, C., Bergado, D.T. and Semprich, S. (2013). Prestressed reinforced soil by geosynthetics—
concept and experimental investigations, Geotextiles and Geomembranes, 37, 109–123.
Lovisa, J., Shukla, S.K. and Sivakugan, N. (2010). Behaviour of prestressed geotextile-reinforced sand
bed supporting a loaded circular footing, Geotextiles and Geomembranes, 28 (2010) 23–32.
Madhavilatha, G. and Somwanshi, A. (2009). Bearing capacity of square footings on geosynthetic rein-
forced sand, Geotex tiles and Geomembranes, 27 (2009) 281–294.
Shivashankar, R., Madhav, M.R. and Miura, N. (1993). Rein forced granular beds overlying soft clay,
Proceedings of 11th South East Asian Geotechnical Conference, Singapore, 409–414.
Shivashankar, R. and Reddy, A.C.S. (1998). Reinforced granular bed on poor filled up shedi ground,
Proceedings of the Indian Geotechnical Conference – 1998, Vol. 1, 301–304.
Shivashankar, R. and Setty, K.R.N.S. (2000). “Foundation problems for ground level storage tanks in
and around Mangalore”, Proceedings of Indian Geotechnical Conference 2000, IIT Bombay, Mumbai.
Unnikrishnan and Aparna Sai. (2014), “Footings in Clay soil supported on Encapsulated Granular
Trench”, Proceedings of Indian Geotechnical Conference 2014, Kakinada.
Vinod, P., Bhaskar, A.B. and Sreehari, S. (2009). Behaviour of a square model footing on loose sand
reinforced with braided coir rope, Geotextiles and Geomembranes, 27 (2009) 464–474.
252

An artificial neural network-based model for predicting the

bearing capacity of square footing on coir geotextile reinforced soil
Dharmesh Lal, N. Sankar & S. Chandrakaran

Department of Civil Engineering, NIT Calicut, Calicut, India
ABSTRACT: Geosynthetic materials like geotextiles, geogrids and geocells have gained
widespread acceptance over recent years due to their superior engineering characteristics
and quality control. The rising cost and the environmental concerns created by these syn-
thetic reinforcement materials makes it necessary to explore alternate resources for soil rein-
forcement. Coir is an eco-friendly, biodegradable, organic material which has high strength,
stiffness and durability characteristics compared to other natural reinforcement materials.
This paper deals with a systematic series of plate load tests on unreinforced sand and sand
reinforced with coir geotextiles. A significant enhancement in strength and stiffness charac-
teristics was obtained with the provision of coir reinforcement. Based on the test results, an
Artificial Neural Network (ANN) model has also been established for predicting the strength
of sand beds when these reinforcement elements are applied practically. The predicted values
from the model and those obtained from the experimental study are found to have a good
correlation.
Keywords: Coir geotextile, artificial neural network, bearing pressure, subgrade stabilization
1 INTRODUCTION
In recent years, geosynthetics have become increasingly popular for their use as a reinforce-
ment in earth structures. The use of geosynthetics in reinforcing sand beds has been stud-
ied by various researchers (Abu-Farsakh et al., 2013; Guido et al., 1986; Akinmusuru &
Akinbolade, 1981; Omar et al., 1993; Ghosh et al., 2005; Latha & Somwanshi, 2009; Sharma
et al., 2009). Synthetic fibers have a longer life and do not generally undergo biological deg-
radation, thus minimizing environmental concern. Coir geotextiles can be considered as an
efficient replacement for their synthetic counterparts due to their economy and excellent
engineering properties. The use of coir as a reinforcement material has been studied by vari-
ous researchers (Lekha, 1997; SivakumarBabu et al., 2008; Vinod et al., 2009; Subaida et al.,
2008). India is one of the leading coir producing countries. While the world focus is shifting
to natural geotextiles, India as a producer of coir geotextiles, has much to gain by using it for
meeting domestic as well as global demands. Natural geotextiles are becoming increasingly
popular in various geotechnical applications like construction of embankments, subgrade
stabilization, slope protection work, weak soil improvement and so on. From the studies
reported so far, it is perceived that the potential of coir products as a reinforcement material
is under-utilized.
Empirical models based on Artificial Neural Networks (ANN) have been widely used
for numerous applications in geotechnical engineering. Neural networks have proved to be
an efficient tool to predict the behavior of soils under different test conditions, especially,
since the relationship between the input and output variables is complex. Although models
253

with different input and output variables are required to be created for the different test
conditions, the process is comparatively easy and more realistic than numerical models. The
present study deals with the formulation of empirical models based on artificial neural net-
works for predicting the behavior of sands reinforced with coir geotextiles, under plate load
testing conditions.
2 ARTIFICIAL NEURAL NETWORK (ANN)
An Artificial Neural Network (ANN) is a form of artificial intelligence, which tries to simu-
late the behavior of the human brain and nervous system. In recent years, many researchers
have investigated the use of artificial neural networks in geotechnical engineering applica-
tions and have obtained reassuring results (e.g. Kung et al., 2007; Kuo et al., 2009; Ornek
et al., 2012; Harikumar et al., 2016). MATLAB software was used for formulating the ANN
model. The parameters used for creating the model are listed in Table 1.
Table 1. Parameters used for creating model.
Settlement level (s/B) 15%
Reinforcement parameters a) Depth to the first layer of

used for modeling reinforcement (u/B)
b) Width of reinforcement
c) Number of reinforcement
layers
Training function Gradient descent with
momentum and adaptive
learning rate back
propagation technique
Number of nodes in the input, 3, 10 and 1, respectively
hidden and output layer
Learning rate 0.01
Momentum constant 0.9
Figure 1. Photograph of coir geotextile used for the study.
254

Figure 2. Photograph of test set up.
3 MATERIALS AND METHODOLOGY
The sand used for the tests had a specific gravity of 2.65, effective size 0.32, coefficient of
uniformity 2.56, coefficient of curvature 0.88 and angle of friction of 38.5° (at 60% relative
density). Classification according to the Unified Soil Classification System (USCS) is SP
(poorly graded sand). All tests were done at a relative density of 60% to simulate medium
dense condition. Figure 1 shows a photograph of the coir geotextiles used for the study. The
properties of woven coir geotextiles were determined as per Indian standards (IS: 13162,
1992; IS: 14716, 1999).
The model test was conducted on a steel tank 750 mm × 750 mm × 750 mm. The model
footing was a square steel plate 150 mm × 150 mm, with a thickness of 25 mm. A hand oper-
ated hydraulic jack was used for loading the footing and a pressure gage of 100 kN was fitted
to measure the load applied. Figure 2 shows a photograph of the test set up. The objectives
of the tests were to study the influence of coir geotextiles on improving the overall perform-
ance of the sand foundations. The test series included varying the depth of the reinforcement
layer from the top of footing (u). An artificial neural network model has also been established
based on the test results. MATLAB software was used to formulate the ANN model. The
technique of formulating empirical models is reliable and relatively easy compared with a
numerical study.
Figure 3 shows the variation of measured values with the values predicted by the model.
From the figure, it can be seen that the bearing capacity predicted by the model and those
obtained from the experimental study are in good agreement.
It can be further seen that the provision of coir reinforcement enhances the strength char-
acteristics of reinforced soil (an almost 95% increase in strength can be observed, even with a
single layer of reinforcement). Additionally, maximum improvement was observed when the
reinforcement was provided at a depth of 0.25 times the width of the foundation (0.25 B).
Furthermore, the predicted and observed values were found to have a good correlation (see
Figure 4), thus establishing the validity of the proposed model.
255

Figure 3. Bearing pressure vs. placement depth (u/B).
Figure 4. Measured vs. predicted bearing capacity.
5 CONCLUSIONS
A detailed, comprehensive study on the performance of coir geotextiles was conducted using
a plate load testing apparatus. The results demonstrated that the coir reinforcement inclu-
256

sions increase the strength and deformation characteristics of sand. An Artificial Neural Net-
work (ANN) model has also been established for predicting the strength of sand beds when
these reinforcement elements are applied practically. The predicted values from the model
and those obtained from the experimental study were found to have a good correlation.
REFERENCES
Abu-Farsakh, M., Chen, Q. & Sharma, R. (2013). An experimental evaluation of the behavior of foot-
ings on geosynthetic-reinforced sand. Soils and Foundations, 53(2), 335–348.
Akinmusuru, J.O. & Akinbolade, J.A. (1981). Stability of loaded footings on reinforced soil. Journal of
Geotechnical Engineering Division, ASCE, 107, 819–827.
Ghosh, A., Ghosh, A. & Bera, A.K. (2005). Bearing capacity of square footing on pond ash reinforced
with jute-geotextile. Geotextiles and Geomembranes, 23(2), 144–173.
Guido, V.A., Chang, D.K. & Sweeney, M.A. (1986). Comparison of geogrid and geotextile reinforced
earth slabs. Canadian Geotechnical Journal, 23, 435–440.
Harikumar, M., Sankar, N. & Chandrakaran, S. (2016). Behavior of model footing on sand bed rein-
forced with multi-directional reinforcing elements. Geotextiles and Geomembranes, 44, 568–578.
IS: 13162. 1992. Geotextiles—methods of test for determination of thickness at specified pressure. New
Delhi: Bureau of Indian Standards.
IS: 14716. 1999. Geotextiles—determination of mass per unit area. New Delhi: Bureau of Indian
Standards.
Kung, G.T.C., Hsiao, E.C.L., Schuster, M. & Juang, C.H. (2007). A neural network approach to esti-
mating deflection of diaphragm walls caused by excavation in clays. Computers and Geotechnics,
34(5), 385–396.
Kuo, Y.L., Jaksa, M.B., Lyamin, A.V. & Kaggwa, W.S. (2009). ANN-based model for predicting the
bearing capacity of strip footing on multi-layered cohesive soil. Computers and Geotechnics, 36(3),
503–516.
Latha, G.M. & Somwanshi, A. (2009). Bearing capacity of square footings on geosynthetic reinforced
sand. Geotextiles and Geomembranes, 27( 4), 281–294.
Lekha, K.R. (1997). Coir geotextiles for erosion control along degraded hill slopes. Proceedings of Semi-
nar on Coir Geotextiles, Coimbatore, India.
Omar, M.T., Das, B.M., Puri, V.K. & Yen, S.C. (1993). Ultimate bearing capacity of shallow founda-
tions on sand with geogrid reinforcement. Canadian Geotechnical Journal, 30, 545–549.
Ornek, M., Laman, M., Demir, A. & Yildiz, A. (2012). Prediction of bearing capacity of circular foot-
ings on soft clay stabilized with granular soil. Soils and Foundations, 52(1), 69–80.
Sharma, R., Chen, Q., Abu-Farsakh, M. & Yoon, S. (2009). Analytical modelling of geogrid reinforced
soil foundation. Geotextiles and Geomembranes, 27, 63–72.
SivakumarBabu, G.L., Vasudevan, A.K. & Sayida, M.K. (2008). Use of coir fibres for improving the
engineering properties of expansive soils. Journal of Natural Fibers, 5(1), 1–15.
Subaida, E.A., Chandrakaran, S. & Sankar, N. (2008). Experimental investigations on tensile and pull-
out behaviour of woven coir geotextiles. Geotextiles and Geomembranes, 26(5), 384–392.
Vinod, P., Ajitha, B. & Sreehari, S. (2009). Behavior of a square model footing on loose sand reinforced
with braided coir rope. Geotextiles and Geomembranes, 27, 464–474.
257

Interference effect of adjacent footings resting on granular beds

overlying weak soils
Aqila Abdul Khader

Department of Civil Engineering, Marian Engineering College, Trivandrum, Kerala, India
Jayamohan Jayaraj
LBS Institute of Technology for Women, Trivandrum, Kerala, India
S.R. Soorya
Department of Civil Engineering, Marian Engineering College, Trivandrum, Kerala, India
ABSTRACT: Due to limited space available for the construction of structures and support
of heavy loads, foundations are often placed close to each other; the footings interact with each
other and their behavior is thus not dissimilar to that of a single isolated footing. This study
aims to determine experimentally the effect of interference of closely spaced shallow footings
(strip footings), resting on granular beds overlying a ‘weak soil’. The laboratory model tests
were carried out at different ‘center to center’ spacing between the footings. The ultimate bear-
ing capacity of footings increased up to a certain critical spacing and thereafter decreased. The
bearing capacity of interfering footings improved due to the provision of a granular bed.
Keywords: laboratory model tests, interference effect, critical spacing, bearing capacity,
granular bed
1 INTRODUCTION
In urban areas, due to the limited space available, foundations often are placed close to each other
resulting in an interference with each other. The interference of the failure zones of the footings
alters the bearing capacity and load-settlement characteristics of the closely spaced footings.
Stuart (1962) was the first pioneer who investigated exclusively the effect of interference of
closely spaced strip footings on ultimate bearing capacity. Using the limit equilibrium technique,
he indicated that the interference of two footings on sand leads to an increase in their ultimate
bearing capacity. Also, he demonstrated that there existed a certain critical spacing between
two footings for which the ultimate bearing capacity becomes maximum. The behavior of the
interference effect is attributed to the phenomenon called the ‘blocking effect’ or ‘arching effect’.
According to this phenomenon, the soil between the two footings forms an inverted arch, and the
combined system of soil and two footings moves down upon loading as a single unit. Since the
area of this single unit is greater than that of the sum of the areas of two footings, it results in
greater bearing capacity. On the other hand, Stuart stated that the interference effect of adjacent
footings on clay would act differently to that on sand, and concluded that the interference effect
would not exhibit any change in bearing capacity as the spacing between the footings decreased.
Selvadurai and Rabbaa (1983) studied the contact stress distribution beneath two inter-
fering rigid strip footings of equal width, resting in frictionless contact with a layer of
dense sand underlaid by a smooth, rigid base. The study showed that the contact stress
distribution for a single isolated foundation has a symmetrical shape and as the spacing
between the adjacent footings decreases, the contact stress distribution exhibits an asym-
metrical shape. Das and Larbi-Cherif (1984) conducted laboratory model tests on two
259

closely spaced strip foundations on sand and concluded that the ultimate bearing capacity
increased up to a critical spacing, but settlement increased due to the interference effect.
Graham et al. (1984) used the method of stress characteristics to calculate the bearing
capacity of a series of parallel footings and thesewere compared with the laboratory tests
conducted on three parallel surface footings resting on sand. Efficiencies were highest at
the closer spacing, and it decreased until the footings behaved independently. Kouzer and
Kumar (2008) examined the interference effect due to equally spaced multiple strip foot-
ings on cohesionless soils, on the ultimate bearing capacity with the help of upper bound
limit analysis in conjunction with finite elements and linear programming, and con-
cluded that the efficiencies were higher at closer spacings between the footings. Kumar
and Bhoi (2009) experimentally studied the interference effect of two closely spaced strip foot-
ings on sand without having any provision of tilt, and found that the bearing capacity of the
footingreached its maximum at a certain critical spacing between the footings. The interference
effect is even more extensive for higher relative densities of sand. Ghosh and Sharma (2010) used
the theory of elasticity to model the settlement behavior on layered cohesionless soil beds and
concluded that the settlement of closely spaced footings was higher than that of a single isolated
footing, which further decreased with the increase in the spacing between the footings. Reddy
et al. (2012) conducted a series of model tests using square and circular footings resting on
sand to study the interference effect, and concluded that bearing capacity and settlement
increased with a decrease in spacing between the footings. Pusadkar et al. (2013) evaluated
the influence of interference of symmetrical footings (square, circular and rectangular foot-
ings) on the bearing capacity of sandy soil and concluded that the bearing capacity improved
with a decrease in the spacing between the footings, whereas, the settlement of the footing
increased. Desai and Moogi (2016) investigated the interference of two strip footings on
three types of soil (soft clay, sandy clay and medium sand) using PLAXIS-2D and found
similar results.
This paper depicts the results of a series of laboratory scale modeling carried out to determine
the interference effect of closely spaced strip footings on granular bed overlying clayey soil.
2 LABORATORY MODEL TESTS
2.1 Materials
For the laboratory tests, well-graded medium sand was used for the granular bed, and locally
available clay was used for a ‘weak soil’. The properties of soils are presented in Table 1 and
Table 2.
Table 1. Properties of the clay used asthe weak soil.
SI No. Property Value
1 Specific gravity 2.5

2 Liquid limit (%) 58
3 Plastic limit (%) 22
4 Shrinkage limit (%) 16.16
5 Percentage of clay (%) 67.2
6 Percentage of silt (%) 30.095
7 Plasticity index (%) 36
8 Permeability (m/s) 3.03 × 10-6
9 UCC strength (kg/cm2) 1.428
10 Friction angle (Φ) 5°
11 Cohesion (kPa) 25
12 Soil classification CH
13 Water content during model test (%) 25
14 Average dry unit weight during model test (kN/m3) 14.6
260

Table 2. Properties of the sand used as the granular bed.
SI No. Property Value
1 Specific gravity 2.66

2 Effective grain size for D10 (mm) 0.208
5 Coefficient of uniformity (Cu) 5.16
6 Coefficient of curvature (Cc) 1.062
7 Permeability (m/s) 1.07 × 10-4
8 Angle of internal friction (Φ) 31.2o
9 Cohesion (c in kPa) 0
10 Soil classification well-graded medium sand
11 Water content during model test (%) 0
12 Average dry unit weight during model test (kN/m3) 19.59
13 Void ratio during model test 0.33
Figure 1. Schematic representation of the test setup.
2.2 Test setup

The test setup consisted of a masonry tank of size 750 mm × 750 mm × 1000 mm assembled
on a loading frame. The model strip footings were of size 50 mm × 50 mm × 670 mm. The
bottom of the footings were made smooth. Also, a loading frame was fabricated to load the
two footings simultaneously. The schematic representation of the test setup is as shown in
Figure 1. The load was applied by a hand operated mechanical jack of 50 kN capacity sup-
ported against a reaction frame. The load was measured using a proving ring and deforma-
tion using two dial gauges placed diametrically opposite to each other. Figure 2 shows the
photographic view of the test setup.
2.3 Preparation of the test bed and testing procedure

First, the tank was filled to the required level with the clay representing weak soil. To achieve
the desired soil density, a layered filling technique was used. The predetermined density of the
clay was used to calculate the weight of the soil required to fill the tank in layers of 50 mm in
height. The sand was added similarly and up to a thickness equal to the width of the footing
(50 mm). Adjacent model footings were placed at varying spacings. The load was applied in
equal increments, and each load increment was maintained until the settlement stopped. The
loading was continued until failure. The Center to Center Spacing (S) between the footings
261

Figure 2. Photographic view of the test setup.
Table 3. Experimental program.
Phases Type Spacing to footing width ratio (S/B)
A Clay (weak soil)

B Granular bed (GB) 1, 1.5, 2, 2.5, 3, 3.5, 4
of Width (B) was expressed as a Spacing to Footing Width Ratio (S/B). The experimental
program was as shown in Table 3.
3.1 Load-settlement behavior

Laboratory scaled model tests were conducted on two symmetrical strip footings of width 50 mm
resting on a granular bed overlying a weak soil. Vertical Stress Versus Normalized Settlement
(δ/B), and settlement to footing width ratio curves for two model strip footings resting on clay at
various spacings are presented in Figure 3. The behavior of a single footing is also presented for
comparison. It is seen from Figure 3? that the closer spacing of footings adversely affected the
load-settlement behavior. The load carrying capacity improved when the spacing was increased
up to three times the width of the footing, indicating that the blocking effect was felt up to
S/B = 3. The plot in Figure 3 highlights the point at which critical spacing occurs. The critical
spacing is defined as the spacing at which maximum bearing capacity occurs. As the spacing
between the footings was increased past the critical spacing, the bearing capacity decreased.
In Figure 4, which presents the vertical stress versus normalized settlement curves for foot-
ings resting on a granular bed overlying clay, a similar behavior was observed. The load-
settlement behavior improved until the spacing was increased up to S/B = 2.5 and until which
the blocking effect had been felt.
Test results showed that by the provision of a granular bed, the interference effect between
the adjacent strip footings was enhanced. Figure 5 shows the improvement in bearing capac-
ity of interfering footings at a spacing of S/B = 2.5.
262

Figure 4. Vertical stress versus normalized
Figure 3. Vertical stress versus normalized settle- settlement curves for footings resting on a
ment curves for footings resting on clay. granular bed overlying clay.
Figure 5. Vertical stress versus normalized

settlement curves of phase A and phase B
at S/B = 2.5.
Figure 6. Variation of interference factor

with S/B at different footing settlements:
Figure 7. Variation of bearing capacity (a) interference effect on weak soil; and (b)
ratio with S/B at different footing settlements. interference effect on a granular bed.
3.2 Interference Factor (IF) and Bearing Capacity Ratio (BCR)

To evaluate the influence of the interference effect of adjacent strip footings and benefi-
cial effects of the provision of a granular bed, two non-dimensional parameters were used
namely: the Interference Factor (IF); and the Bearing Capacity Ratio (BCR).
IF is defined as: the ratio of average pressure on interfering footings of a given size, at a
given magnitude of settlement, to the average pressure on an isolated footing of the same size
and at the same magnitude of settlement.
BCR is defined as: the ratio of average pressure on an isolated or interfering footing of
a given size at a given magnitude of settlement on the granular bed overlying weak soil, to
the average pressure on an isolated or interfering footing of the same size and at the same
magnitude of settlement on the weak soil. The IF and BCR were defined for different footing
settlements at s/B = 5%, 7% and 9% (Kumar & Bhoi, 2009).
263

The variation of IF with S/B associated with different footing settlements for both phase A
and phase B is shown in Figures 6a and 6b. It can be noticed that in all cases the value of IF
becomes maximum at a critical spacing of S/B = 3 in the case of weak soil, and at S/B = 2.5
in the case of granular bed overlying weak soil.
The variation of BCR with S/B associated with different footing settlements is shown in
Figure 7. The maximum BCR is obtained at a spacing of S/B = 1.5.
3.3 Comparison with available literature

A large amount of research has been conducted on the interference effect and bearing capac-
ity of closely spaced footings on sand. Across the majority of the research, it was concluded
that when two footings have no clear spacing between them, the footings will act as a single
footing with twice the width.
As the spacing between footings increased, the bearing capacity of the footings increased,
up to a certain spacing where the interference effect no longer played a part and the bearing
capacity decreased. The trend displayed in Figure 3 and Figure 4 is very similar to the trend
discussed in available literature on the interference effect of two footings on sand.
4 CONCLUSIONS
In the present study, a series of laboratory scaled model tests were performed to determine
the load-settlement behavior of interfering strip footings resting on a granular bed overlying
a weak soil. The following conclusions are drawn:
• Interference of adjacent footings significantly affects the load-settlement behavior. There is an
optimum distance between the footings at which the bearing capacity has the maximum value
due to a blocking effect. In footings resting on clay, the blocking effect appears at S/B = 3, and
for footings resting on the granular bed, the blocking effect appears at S/B = 2.5. However, the
critical value of S/B at which IF becomes the maximum is not a fixed.
• The trend obtained from the laboratory study is similar to the trend discussed in the litera-
ture review on the interference effect of two footings on sand.
• The bearing capacity of closely spaced strip footings improved with the provision of gran-
ular bed over a weak soil.
REFERENCES
Das, B.M. & Larbi-Cherif, S. (1984). Bearing capacity of two closely-spaced shallow foundations on sand.
Desai, M.V.G. & Moogi, V.V. (2016). Study of interference of strip footing using PLAXIS-2D. Interna-
tional Advanced Research Journal in Science, Engineering and Technology, 3(9), 13–17.
Ghosh, P. & Sharma, A. (2010). Interference effect of two nearby strip footings on layered soil: Theory
of elasticity approach. Acta Geotechnica, 5, 189–198.
Graham, J., Raymond, C.P. & Suppiah, A. (1984). Bearing capacity of three closely-spaced footings on
sand. Geotechnique, 34(2), 173–182.
Kouzer, K.M. & Kumar, Jyant. (2008). Ultimate bearing capacity of equally spaced multiple strip
footings on cohesionless soils without surcharge. International Journal for Numerical and Analytical
Methods in Geomechanics, 32, 1417–1426.
Kumar, J. & Bhoi, M.K. (2009). Interference of two closely spaced strip footings on sand. Journal of
Geotechnical and Geoenvironmental Engineering, 135(4), 595–604.
Pusadkar, S.S., Gupta, R.R. & Soni, K.K. (2013). Influence of interference of symmetrical footings on
bearing capacity of soil. International Journal of Engineering Inventions, 2(3), 63–67.
Reddy, S.E., Borzooei, S. & Reddy, N.G.V. (2012). Interference between adjacent footings on sand.
International Journal of Advanced Engineering Research and Studies, 1(4), 95–98.
Selvadurai, A.P.S. & Rabbaa, S.A.A. (1983). Some experimental studies concerning the contact stresses
beneath interfering rigid strip foundations resting on a granular stratum. Canadian Geotechenical-
Journal, 20, 406–415.
Stuart, J.G. (1962). Interference between foundations, with special reference to surface footings in sand.
Geotechnique, 2(1), 15–22.
264

Influence of shape of cross section of footing on load-settlement

behaviour
B. Anusha Nair, Akhila Vijayan, S. Chandni, Shilpa Vijayan, J. Jayamohan & P. Sajith
ABSTRACT: In general the shape of cross section of footings provided for structures is rec-
tangular. There are well accepted theories to determine the bearing capacity and settlements
of footing with a flat base. By altering the cross sectional shape of the footing, better confine-
ment of underlying soil can be attained thereby improving bearing capacity and reducing set-
tlements. In this investigation a series of finite element analyses are carried out to determine
the influence of shape of cross section of the footing on the load-settlement behaviour of strip
footings. It is observed that by altering the shape of cross section of footings, better confine-
ment of underlying soil can be achieved, thereby improving the load-settlement behaviour.
Keywords: Shape, Cross section, Footing, Load-Settlement behaviour
1 INTRODUCTION
The foundation transmits the load of the structure safely to the ground, without under-
going any shear failure or excessive settlement. The bearing capacity of footings has been
extensively studied, both theoretically and experimentally, over the past many decades. The
theoretical approach was initiated by Prandtl (1921) and Reissner (1924), and the design-
oriented bearing capacity equation (fully considering the soil unit weight, cohesion, and
friction angle) was proposed by Terzaghi (1943). After the early development of the bear-
ing capacity solution, most efforts have focused mainly on a more realistic derivation of
Table 1. Cross sectional shapes of footings.
Shape of footing d/B
Footing with rectangular cross section –
Rectangular cross section with flanges 0.2, 0.4, 0.6, 0.8
Sloped cross section 0.2, 0.4, 0.6, 0.8
265
CH045_199.indd 265 7/9/2018 5:23:50 PM

bearing capacity and various correction factors [Meyerhof G. (1965), Taiebat and Carter
(2000), Griffiths et al. (2002), Ericson and Drescher (2002), Michalowski and Dawson (2002),
Salgado et al. (2004), Lee and Salgado (2005)].
The vertical bearing capacity of a shallow foundation is a classical geotechnical problem.
The mechanism of soil failure transforms from a general shear failure for a surface footing
(Craig 2004) to a localized failure for a buried footing (Hu et al. 1999; Wang and Carter
2002; Hossain and Randolph 2009, 2010; Zhang et al. 2012, 2014), with the ultimate bearing
capacity for a deeply buried foundation demonstrated to be considerably larger than that of a
shallow foundation (Merifield et al. 2001; Zhang et al. 2012). Ming and Radoslaw (2005) pro-
posed new shape factors for square and rectangular footings based on elasto-plastic model of
the soil and finite element analyses.
Singh and Monika (2016) carried out studies on Shell foundations of various shapes of
cross section. In all the types of footings; except shell foundations, the shape of cross sec-
tion is rectangular and the base has been considered as a plane surface. The pattern of soil
movement beneath the footing during loading is a significant factor contributing to load-
settlement behaviour. By altering the shape of cross section of footing, it would be possible to
provide better confinement of underlying soil thus improving bearing capacity.
The purpose of this paper is to investigate numerically the influence of shape of cross sec-
tion on the load-settlement behaviour. A series of non-linear finite element analyses are car-
ried out with the FE software PLAXIS 2D. Three different shapes of cross section of footing,
shown in Table 1, are considered for the analyses.
Property Clay
Dry Unit Weight (kN/m3) 16

Specific Gravity 2.1
Cohesion (kPa) 20
Angle of Shearing Resistance (°) 0
Modulus of Elasticity (kPa) 7000
Poisson’s Ratio 0.3
266

obtained from direct shear tests; internal friction angle and cohesion intercept. Since strip
footing is used, plane strain model is adopted in the analyses. The settlement of the rigid
15 noded triangular elements. Mesh generation can be done automatically. Medium mesh size
is adopted in all the simulations.The size of the strip footing (B) is taken as one metre and the
width and depth of soil mass are taken as 10B in all analyses The footing and the confining
walls are modelled using plate elements. To simulate the interaction between the footing and
underlying soil, an interface element is provided at the bottom surface of footing. The geomet-
ric model is shown in Fig. 1 and the typial stress distribution after loading is shown in Figure 2.
The soil is modeled using 15-node triangular elements. Properties of locally available sand
and clay are adopted in the analyses. The material properties adopted are outlined in Table 2.
Vertical Stress vs Settlement curves for Rectangular Cross Section with flanges; obtained
from finite element analyses are presented in Figure 3. It is seen that the presence of flanges
below the edges of the strip footing improves the load-settlement behaviour. The improve-
ment for d/B values up to 0.6 is less whereas for values of d/B > 0.6, improvement is more.
The load-settlement behaviour of footing with sloping cross section is presented in
Figure 4. The optimum improvement is observed when d/B = 0.4. For higher values of d/B,
Figure 2. Typical stress distribution after loading.
Figure 3. Vertical stress vs settlement curves for rectangular cross section with flanges.
267

Figure 4. Vertical stress vs settlement curves for footing with sloped cross section.
Figure 5. Improvement factor.
the improvement reduces. This behaviour is quite contrast to that of Rectangular Cross Sec-
tion with Flanges.
To quantify the improvement in load-settlement behaviour attained due to anchoring
the reinforcement, an Improvement Factor (If) is defined as the ratio of stress with Various
Shapes of Cross Section to that with Rectangular Cross Section; at 0.5 mm settlement.
It is seen that for rectangular cross section with flanges, the improvement factor initially
increases and then reduces when d/B = 0.6. However for further higher values of d/B, the
improvement factor increases.
The optimum value of d/B for footing with sloping cross section is observed to be 0.4. For
higher values of d/B, the improvement factor is reducing.
4 CONCLUSIONS
From the finite element analyses carried out, the following conclusions are drawn.
• The load-settlement behaviour can be improved by altering the shape of cross section of
footings
• The improvement factor depends of the geometrical parameters of the cross section
268

REFERENCES
Craig, R.F. (2004). Craig’s soil mechanics, Taylor & Francis, New York.
Ericson Hans L, Drescher Anderw. Bearing capacity of circular footings. J Geotech Geoenviron Eng,
ASCE 2002; 128(1):38–43.
Griffiths D.V., Fenton Gordon A., Manoharan N. Bearing capacity of rough rigid strip footing on
cohesive soil: probabilistic study. J Geotech Geoenviron Eng, ASCE 2002; 128(9):743–50.
Hossain, M.S., and Randolph, M.F. (2009). “New mechanism-based design approach for spudcan foun-
dations on single layer clay.” J. Geotech. Geoenviron. Eng., 10.1061/(ASCE)GT.1943-5606.0000054,
1264–1274.
Hossain, M.S., and Randolph, M.F. (2010). “Deep-penetrating spudcan foundations on layered clays:
centrifuge tests.” Géotechnique, 60(3), 157–170.
Hu, Y., Randolph, M.F., and Watson, P.G. (1999). “Bearing response of skirted foundation on non
homogeneous soil.” J. Geotech. Geoenviron. Eng., 10.1061/(ASCE)1090-0241 (1999), 125:11(924),
924–935.
Lee J. and Salgado R. Estimation of bearing capacity of circular footing on sands based on CPT. J
Geotech Geoenviron Eng, ASCE 2005; 131(4):442–52.
Merifield, R.S., Sloan, S.W., and Yu, H.S. (2001). “Stability of plate anchors in undrained clay.”
Géotechnique, 51(2), 141–153.
Meyerhof G. Shallow foundations. J Soil Mech Found Div, ASCE 1965; 91(SM2):21–31.
Michalowski Randoslaw L. and Dawson M.E. Three-dimensional analysis of limit loads on Mohr–
Coulomb soil. Found Civil Environ Eng 2002; 1(1):137–47.
Ming Zhu and Radoslaw L. Michalowski, (2005), “Shape Factors for Limit Loads on Square and
Rectangular Footings”, Journal of Geotechnical and Geoenvironmental Engineering, Vol. 131, No. 2,
ASCE, 223–231.
Prandtl L. Über die Eindringungsfestigkeit (Härte) plastischer Baustoffe und die Festigkeit von Schnei-
den. Zeit Angew Math Mech 1921(1):15–20.
Reissner H. Zum Erddruckproblem. In: Proceedings of first international congress of applied mechan-
ics. Delft; 1924. p. 295–311.
Salgado R., Lyamin A.V., Sloan S.W., Yu H.S. Two-and three-dimensional bearing capacity of founda-
tion in clay. Geotechnique 2004; 54(5):297–306.
Singh, S.K., and Monika, K., (2016), “Load carrying capacity of shell foundations on treated and
untreated soils” Indian Geotechnical Conference 2016, IIT Madras Taiebat HA, Carter JP. Numeri-
cal studies of the bearing capacity of shallow foundation. Geotechnique 2000; 50(4):409–18.
Terzaghi K. Theoretical soil mechanics. New York: Wiley; 1943.
Wang, C.X., and Carter, J.P. (2002). “Deep penetration of strip and circular footings into layered clays.”
Int. J. Geomech.,10.1061/(ASCE)1532-641, (2002), 2:2(205), 205–232.
Zhang, Y., Bienen, B., Cassidy, M.J., and Gourvenec, S. (2012). “Undrained bearing capacity of deeply
buried flat circular footings under general loading.” J. Geotech. Geoenviron.Eng., 10.1061/(ASCE)
GT.1943-5606.0000606, 385–397.
Zhang, Y., Wang, D., Cassidy, M.J., and Bienen, B. (2014). “Effect of installation on the bearing capac-
ity of a spudcan under combined loading in soft clay.” J. Geotech.Geoenviron. Eng., 10.1061/(ASCE)
GT.1943-5606.0001126, 04014029.
269

Experimental study to determine the elastic wave velocities of

marineclay
K. Anagha & M.N. Sandeep

Department of Civil Engineering, IES College of Engineering, Thrissur, Kerala, India
K.S. Beena
Department of Civil Engineering, Cochin University of Science and Technology, Kochi, Kerala, India
ABSTRACT: P-wave velocity and shear wave velocity are the dynamic parameters used to
determine soil characteristics. Ultrasonic pulse velocity tests are usually used for p-wave velocity
measurements. In this study, tests were conducted to investigate the effect of water content and
dry densities on p-wave velocity and unconfined compressive strength and thus, to establish a
correlation between p-wave velocity and unconfined compressive strength. Marine clay was the
material used for the study. Experiments were conducted on a soil sample of diameter 3.75 cm
and length 7.5 cm. Results show that p-wave velocity and unconfined compressive strength dem-
onstrate a similar trend while varying the parameters such as water content and dry density. The-
oretical correlations connecting elastic wave velocities and electrical resistivity were established
for finding the shear wave velocities. Gassmann’s and Archie’s equations for porosity were used
to derive the equation for shear wave velocity. Electrical resistivity was calculated experimentally.
1 INTRODUCTION
Elastic waves are generated in nature by the movement of tectonic plates, explosions, land-
slides and so on. Seismic waves are the vibrations generated at the interior of the earth, dur-
ing rupture or explosions, which take energy away from the center of the earth. Vibrations
which travel through the interior of the earth rather than surface are called body waves. Body
waves are classified into two, primary waves or compression waves (p-waves), and second-
ary or shear waves (s-waves). The determination of elastic waves and their movement is an
important parameter for different field applications such as insertion of deep foundations
in soil, soil stabilization, compaction characteristics, anisotropic behavior of soils, stiffness
evaluation of soils, sample quality determination, stratification of soils and so on.
These waves can also be modeled artificially by using different piezoelectric transducers in the
field as well as in the laboratory. P-wave velocity is mainly determined using ultrasonic meth-
ods. Ultrasonic methods are usually used for assessing concrete quality and strength determina-
tion (Lawson et al., 2011). Similarly, ultrasonic methods can also be used for investigating the
compaction characteristics of soils, finding that the variation of velocity with water content is
similar to the variation of density with water content (Nazli et al., 2000). P-wave velocity, shear
wave velocity and damping characteristics can be estimated by different calibrated ultrasonic
equipment (Zahid et al., 2011). Elastic wave velocities are determined by piezo disk elements
and electrical resistivity by electrical resistivity probes. A theoretical correlation connecting the
elastic wave velocity and electrical resistivity was proposed by Jong et al. (2015).
The focus of this study was to find the primary wave velocity of soft soil using an ultra-
sonic pulse velocity test at different water contents and dry densities and to correlate the
primary wave velocities to the unconfined compressive strength of the soil. The shear wave
velocity of the soil was determined by proposing a new equation for shear wave velocity
which connects elastic wave velocity and electrical resistivity.
271

Figure 1. Marine clay used for the study.
Properties Values

Liquid limit (%) 70
Plastic limit (%) 24.4
Plasticity index (%) 45.6
Shrinkage limit (%) 12.7
Clay (%) 47
Silt (%) 33
Sand (%) 20
Max.dry density (g/cc) 1.62
Optimum moisture content (%) 20
2 MATERIALS USED
The soil used for the study was marine clay, which was collected from Kochi. It is blackish in
color. Figure 1 shows the soil used for this study.
The soil properties were determined and are listed in Table 1.
3 TEST DESCRIPTION
To prepare specimens of diameter 3.75 cm and length 7.5 cm for the unconfined compressive
test and ultrasonic pulse velocity test, three densities 1.62 g/cc, 1.5 g/cc and 1.4 g/cc were fixed
from the compaction curve. The compaction curve is shown in Figure 2. Water content cor-
responding to these densities were taken from the wet and dry sides of the compaction curve
and are tabulated in Table 2.
3.1 Ultrasonic pulse velocity test

A pair of piezo electric transducers, which are capable of producing stress waves with frequen-
cies higher than 20 kHz were used for testing. Here, 60 kHz center frequency transducers were
used. A battery controlled ultrasonic tester was used to obtain the time of travel through the
soil. Through transmission method is used for the study, in which two opposite surfaces are
available for testing. After calibrating the instrument with a glass prism, a sample was placed
in between the two identical transducers. A suitable couplant was used on the surfaces of the
transducer to transmit the signal effectively into the specimen. By dividing the path length
or length of specimen by the time required to travel the length, ultrasonic pulse velocity or
p-wave velocity can be calculated. The test setup for the ultrasonic pulse velocity test is shown
in Figure 3.
272

Figure 2. Compaction curve.
Table 2. Sample properties.
Sample Dry density (g/cc) Water content (%)
1 1.62 20
2 1.5 17.5
3 1.5 21.4
4 1.4 15.3
5 1.4 23
Figure 3. Test setup for ultrasonic pulse velocity test.
3.2 Electrical resistivity test

An apparatus was constructed for measuring electrical resistivity, which consists of copper
electrodes, cylindrical mold, ammeter, voltmeter, auto transformer and an AC Supply. It is
grooved to fix a cylinder into it. A cylinder of length 8.7 cm and diameter 3.75 cm was used to
prepare the mold. The cylinder used was a PVC pipe, which is a good electrical insulator. The
mold was filled to the required moisture content and dry density. Then, it was connected to an
alternating current source of 230 V. To regulate the voltage, an auto transformer with varying
voltage between 0–200 V was used. A voltmeter of accuracy 10 V was connected in parallel to
the mold and an ammeter in series to the mold. A voltage of 10 V was applied to the sample
and then current flowing through the sample was measured from the ammeter and resistance
was calculated from Ohm’s Law. These electrical resistance values are used for calculating
resistivity. Figure 4 shows the experimental arrangement for finding the electrical resistivity.
3.3 Shear wave velocity from electrical resistivity values

Archie developed (Jong et al., 2015), an equation for electrical resistivity of soil mixture in
terms of porosity (n) and electrical resistivity of the electrolyte (Eel):
273

Figure 4. Test setup for finding electrical resistivity of soil.
Emix = α Eel n–m (1)
where α is the cementation factor taking values between 0.6 and 3.5, ‘m’ is the shape factor
indicating the shape of the particle, porous structure and specific surface. It takes values in
the range of 1.4–2.2. The only term which connects the electrical resistivity and elastic wave
velocity is the porosity. Movement of waves results volume change in soil. Gassmann sug-
gested (Jong et al., 2015) an equation for bulk modulus in terms of bulk modulus of soil
grain (Bg), bulk modulus of soil skeleton (Bsk), bulk modulus of pore fluid (Bf) and porosity:
 Bsk 
1 − Bg 
BGassmann = Bsk + (2)
n Bsk 1 − n
− +
Bf Bg 2 Bg
Gassmann suggested another equation for bulk density in terms of the elastic wave velocities:
BGassmann = ρ (Vp² -4/3 Vs²) (3)
By equating Equations 2 and 3, an equation for porosity is obtained as:

Bsk
−1
Bg 1 Bsk
− +
 2 4 2  Bg Bg 2
Bsk − ρ Vp − Vs 
 B 
nGassmann = (4)
1 1
−
Bf Bg
From Equation 1:
nArchie = Emix1/–m/α Eel (Jong et al., 2015) (5)
By equating Equations 4 and 5, a new equation for finding the shear wave velocity is
derived.
The equation for shear wave velocity can be written as:
( ( (
Vs = √ 3  √ E mix1/( − m ) ( ρVp 2 − Bsk ) Bg − B f Bg 2
 )
(( )
− Bg − Bsk ρVp 2 + Bg 2 − Bsk 2 × α Eel Bg B f ) )) (6)

(2 √ ρ √((E mix
1/( − m )
(B
g ) ) ( )
− B f Bg − Bg − Bsk α Eel Bg B f
2
))
274

where Bg, Bf, Bsk are the bulk moduli of the soil grain, fluid and the soil skeleton, respectively,
and it takes the values as 14.5 GPa, 2.18 GPa and 7.78 GPa and Eel is the electrical resistivity
of electrolyte and is taken as 0.312 Ωm (Jong et al., 2015).
4.1 P-wave velocity measurements

4.1.1 V
ariation of ultrasonic pulse velocity and unconfined compressive strength with
dry densities
Figures 5 and 6 show the variation of p-wave velocity and unconfined compressive strength
with dry density at dry of optimum and wet of optimum water contents respectively.
For both wet of optimum and dry of optimum water contents, as the dry density increases
both p-wave velocity and unconfined compressive strength increase.
The increase in p-wave velocity for an increase in dry density shows that wave propaga-
tion will be faster in the case of solids than through voids. Both p-wave velocity and uncon-
fined compressive strength show a similar trend while varying dry density for different soils.
Therefore, the ultrasonic pulse velocity can be used as a parameter to estimate unconfined
compressive strength of soil indirectly.
4.1.2 V
ariation of ultrasonic pulse velocity and unconfined compressive strength with varying
water contents
Table 3 shows obtained p-wave velocity and unconfined compressive strength.
As the water content increases from 17.5 to 21.4%, p-wave velocity and the unconfined
compressive strength decreases, for the dry density of 1.5 g/cc. For 1.4 g/cc, water content
increases from 15.3 to 23%, while both p-wave velocity and unconfined compressive strength
decreases. Waves will propagate faster in solids than liquids for the same dry density.
4.2 Shear wave velocity measurements

The shear wave velocity soil is obtained from Equation 6, which connects the electrical resis-
tivity with elastic wave velocities. Table 4 shows the obtained resistivity from the resistivity
method and obtained shear wave velocity by using Equation 6 for soil.
For marine clay, the resistivity values range from 0.86 to 6.23 Ωm. The shear wave velocity
values for marine clay vary between 255 and 337 m/s.
This method can be effectively used to determine the shear wave velocity of soft soils and
it can provide a fast approach for finding the shear wave velocities.
Figure 5. Variation of unconfined compres- Figure 6. Variation of unconfined compres-

sive strength and p-wave velocity for different sive strength and p-wave velocity for different
dry density at dry of optimum water content. dry density at wet of optimum water content.
275

Table 3. Obtained p-wave velocity and unconfined compressive strength.
Dry density (g/cc) 1.4 1.5

Moisture content (%) 15.3 23 17.5 21.4
P-wave velocity (m/s) 320.16 294.11 348.31 314.8
Decrease in 8.14 9.62
velocity (%)
Unconfined compressive strength (kN/m2) 46 7.3 50.4 10.8
Decrease in strength (%) 84.1 78.6
Table 4. Obtained electrical resistivity and shear wave velocity.
Dry density and water content Electrical resistivity (Ωm) Shear wave velocity (m/s) Vs
1.62 g/cc, 20% 6.23 336.58

1.5 g/cc, 17.5% 3.48 301.64
1.5 g/cc, 21.4% 1.53 272.63
1.4 g/cc, 15.3% 2.05 277.26
1.4 g/cc, 23% 0.86 254.71
5 CONCLUSIONS
P-wave velocity and shear wave velocity are important parameters in geotechnical engineer-
ing and can be used to predict the properties of soil without sampling and testing. The ultra-
sonic pulse velocity test is mainly used to determine the p-wave velocity.
As the dry density increases, both compression wave velocity and unconfined compres-
sive strength increase for both dry of optimum and wet of optimum water contents. This is
because the wave transmission through solids is faster than through voids. P-wave velocity
and unconfined compressive strength decrease with an increase in moisture content for the
same dry density. The decrease in p-wave velocity is due to the slower rate of wave propaga-
tion in liquids than in solids. Therefore, the ultrasonic pulse velocity can be used as a param-
eter to estimate unconfined compressive strength of soil indirectly.
A theoretical correlation connecting elastic wave velocities and electrical resistivity was
used for finding the shear wave velocities. It was found that the shear wave velocity of the soil
varies with changes in the dry density and water content of the soil.
REFERENCES
Jong, S.L. & Yoon, M.H. (2015). Theoretical relationship between elastic wave velocity and electrical
resistivity. Journal of Applied Geophysics, 116, 51–61.
Lawson, K.A., Danso, H.C., Odoi, C.A. & Quashie, F.K. (2011). Non-destructive evaluation of con-
crete using ultrasonic pulse velocity. Research Journal of Applied Sciences, Engineering and Technol-
ogy, 3(6), 499–504.
Nazli, Y. Inci, G. & Miller, C.J. (2000). Ultrasonic testing for compacted clayey soils. Journal of Geotech-
nical and Geoenvironmental Engineering, ASCE, 287, 5, 54–68.
Nazli, Y., James, L.H. & Mumtaz, A.U. (2003). Ultrasonic assessment of stabilized soils. Journal of
Geotechnical and Geoenvironmental Engineering, ASCE, 301, 14,170–181.
Zahid, K., Cascante, G. & Naggar, M.H. (2011). Measurement of dynamic properties of stiff specimens
using ultrasonic waves.Canadian Geotechnical Journal, 48, 1–15.
276

Behavior of a single pile under combined and uplift loads:

A review
D. Divya, R.B. Jiniraj & P.K. Jayasree

Department of Civil Engineering, College of Engineering Trivandrum, Thiruvananthapuram, India
ABSTRACT: Pile foundations are deep foundations constructed to transfer loads from the
superstructure to the hard strata beneath. The common loads acting on pile vertical, lateral and
uplift loads. As the vertical load on the pile changes, variations are shown by the pile in its lateral
behaviour. As the vertical load increases, the lateral capacity of the piles also tends to increase.
The lateral behaviour of the pile under combined loading depends on various factors such as
the order of loading, properties of the soil and pile. Even though the vertical force is the most
common force acting on a pile, uplift forces are also seen, especially in foundations of structures
in a harbor and in the case of submerged platforms of waterfront structures. Hydrostatic pres-
sure, overturning moments, lateral force and swelling of surrounding soil cause uplift of piles. In
the case of sandy soils, uplift resistance depends on the skin friction of piles and is determined
by considering the shape of the failure surface, the shear strength of the soil and weight of the
pile. This paper presents a review of the research works conducted on pile foundations subject
to combined (vertical and lateral) loading conditions and foundations subject to uplift loads.
1 INTRODUCTION
Pile foundations are slender members provided to transfer a load from the super structure to
the hard strata beneath. Lateral loads are also experienced in pile foundations of some struc-
tures. High velocity wind, dynamic earthquake pressure, soil pressure and so on, can cause lat-
eral loads. Practically, when lateral loading is present on pile foundations, the actual condition
occurring involves combined vertical and lateral loading on the structures. Hence, the behavior
of piles under combined loading is an important matter to be considered during the structural
design of piles. Many theoretical methods exist to evaluate the behavior of piles under different
loads such asaxial-compressive, axial-uplift and lateral loading. To minimize the complexity
in analyzing different loads simultaneously, loads are analyzed separately. Vertical loads on
piles are analyzed to determine the bearing capacity and lateral loads to determine the elastic
behavior of piles. However, in the field, lateral loads are of a large order, hence the effect of
combined loading cannot be neglected. Foundations of retaining walls, anchors for bulkheads,
bridge abutments, piers, anchorage for guyed structures and offshore structures, supported on
piles are exposed to large inclined uplift loads. Most of the studies on piles are concentrated
on vertical loads rather than uplift forces. Uplift forces may result either from a lateral force or
direct pull out, hydrostatic pressure, overturning moments, lateral force and swelling of the sur-
rounding soil. The effect of geometric properties like slenderness ratio on the lateral capacity
of piles under combined lateral and vertical loading has also been a topic of study. This paper
presents a review on the research works addressing both of these aspects.
2 INFLUENCE OF VERTICAL LOAD ON THE LATERAL RESPONSE OF THE PILE
Combined loading effects on piles have been a matter of study for the last ten decades.
Trochanis et al. (1991) used 3D FE to study the effect of combined loading on piles and
277

reported that a vertical load has a lesser effect on the lateral behavior piles. Anagnostopoulos
and Georgiadis (1993) studied the lateral behavior of piles under axial load using experimen-
tal models and 2D FE analysis, and reported that FEM analysis is a useful tool to predict the
stress developed and volume changes in soil mass under combined vertical and lateral loads
when compared to other theoretical and 2D analytical methods. Karthigeyan et al. (2006,
2007) investigated the performance of piles under combined loading using 3D FE analysis
and noticed that sandy soil experienced an increase in lateral capacity with an increase in
vertical load, while in clayey soil the trend is reversed. Figure 1 shows the increase in lateral
capacity with an increase in vertical load for sandy soil and Figure 2 shows the lateral behav-
ior of piles under vertical load for various slenderness ratios. Achmus and Thieken (2010)
used 3D FE methods to analyze the behavior of piles when acted upon by vertical and lateral
loads in sandy soil and reported that the lateral loads mobilize passive earth pressure whereas
vertical loads mobilize skin friction.
Hussien et al. (2012, 2014a, 2014b) used FE models to study the soil-pile interaction of
free head piles installed in sandy soils and noticed a slight increase in lateral capacity of piles
under vertical loads and inferred that the improvement in lateral capacity of piles may be
due to the increase in confining pressure in the surrounding sand deposits. Maru and Vanza
Figure 1. Effect of vertical load on lateral capacity (Source: Karthigeyan et al., 2007).
Figure 2. The lateral behavior of piles under vertical load for various slenderness ratios (Source:
Karthigeyan et al., 2006).
278

Figure 3. Variation of uplift load capacity of a single pile with L/D ratio (Source: Parekh & Thakare, 2016).
(2017) conducted laboratory tests to investigate the lateral behavior of piles under axial-
compressive loads and reported that the lateral load carrying capacity of piles improved with
an increase in axial load and slenderness ratio. Hazzar et al. (2017) used 3D FD to study the
effect of axial loads on the lateral behavior of pile foundations in sand and concluded that the
behavior of piles in the lateral direction is not affected by the axial loads acting on the piles.
3 INFLUENCE OF UPLIFT LOAD ON PILES
The study of the behavior of a single pile under uplift loads and the factors affecting the
uplift capacity of piles needs to be improved. Ayothiraman and Reddy (2015) investigated
pile behavior under combined uplift and lateral loading in sand. The results concluded that
the uplift load v/s axial displacement behavior is nonlinear for independent loading and in
the case of combined loading. The behavior of a single pile, square pile group and hexago-
nal pile group under independent lateral and uplift loading, and also under combined uplift
and lateral loading were studied by Parekh and Thakare (2016). Based on their results, they
concluded that, with an increase in L/D ratio of the pile, the uplift load capacity of a single
pile increases linearly upto a certain point for both independent uplift loading and combined
loading with a constant lateral load. Figure 3 shows the variation of uplift load capacity
of single piles with an L/D ratio. To determine the ultimate uplift capacity of piles in sand,
Chattopadhyay and Pise (1986) proposed an analytical method with an assumed curved fail-
ure surface through the soil and determined the effects of factors like L/D ratio, the angle of
shearing resistance and pile friction angle on the ultimate uplift capacity of the pile. Das and
Seeley (1975) conducted some model tests in loose granular soil for determining the ultimate
uplift capacity of vertical piles under axial pull. The results include the variation of unit
uplift skin friction with the embedment depth. Rao and Venkatesh (1985) conducted labora-
tory studies in uniform sands to find the uplift behavior of short piles and reported that the
uplift capacity of piles was found to increases with the L/D ratio, the density of soil, particle
size and pile roughness.
4 CONCLUSION
• In non-cohesive soils, the lateral behavior of piles is influenced by the vertical load acting
on them which shows an increase in the lateral load carrying capacity.
• The effect of vertical load depends on the sequence of loading, geometric properties of the
pile and material properties of the soil.
• Lateral load carrying capacity shows an increasing trend with an increase in vertical load
as well as slenderness ratio.
279

• The ultimate uplift capacity of piles increases substantially under uplift and lateral loading.
• With an increase in both L/D ratio and relative density of the soil, the uplift capacity of
piles increases.
• The pile head uplift movement shows a considerable increase under combined loading but
very little on independent uplift and lateral loading.
• Longer piles are more resistant to uplift load compared to shorter piles.
• The resistance of piles under uplift load increases as the angle of shearing resistance and
soil-pile friction angle increase.
REFERENCES
Achmus, M. & Thieken, K. (2010). On the behaviour of piles in non-cohesive soil under combined hori-
zontal and vertical loading. ActaGeotechnica, 5(3), 199–210.
Anagnostopoulos, C. & Georgiadis, M. (1993). Interaction of axial and lateral pile responses. Journal of
Geotechnical Engineering, 119(4), 793–798.
Ayothiraman, R. & Reddy, K.M. (2015). Experimental studies on behavior of single pile under com-
bined uplift and lateral loading. Journal of Geotechnical and Geo environmental Engineering, ASCE,
141, 1–10.
Chattopadhyay, B.C. & Pise, P.J. (1986). Uplift capacity of piles in sand. Journal of Geotechnical Engi-
neering, ASCE, 112(9), 888–904.
Das, B.M. & Seeley, G.R. (1975). Uplift capacity of buried model piles in sand. Journal of GTE Div.
ASCE, 10, 1091–1094.
Hazzar, L., Hussien, M.N. & Karray, M. (2017). Influence of vertical load on lateral response of pile
foundation in sands and clays. Journal of Rock Mechanics and Geotechnical Engineering, 9, 291–304.
Hussien, M.N., Tobita, T., Iai, S. & Karray, M. (2014a). Influence of pullout loads on the lateral
response of pile foundation. Proceedings of the 67th Canadian Geotechnical International Conference.
Hussien, M.N., Tobita, T., Iai, S. & Karray, M. (2014b). On the influence of vertical loads on the lateral
response of pile foundation. Computers and Geotechnics, 55, 392–403.
Hussien, M.N., Tobita, T., Iai, S. & Rollins, K.M. (2012). Vertical load effect on the lateral pile group
resistance in sand response. Geomechanics and Geoengineering, 7(4), 263–282.
Karthigeyan, S., Ramakrishna, V.V.G.S.T. & Rajagopal, K. (2006). Influence of vertical load on the
lateral response of piles in sand. Computers and Geotechnics, 33, 121–131.
Karthigeyan, S., Ramakrishna, V.V.G.S.T. & Rajagopal, K. (2007). Numerical investigation of the effect
of vertical load on the lateral response of piles. Journal of Geotechnical and Geoenvironmental Engi-
neering, 133(5), 512–521.
Maru, V. & Vanza, M.G. (2017). Lateral behaviour of pile under the effect of vertical load. Journal of
Information, Knowledge and Research in Civil Engineering, 4(2), 482–485.
Parekh, B. & Thakare, S.W. (2016). Performance of pile groups under combined uplift and lateral
loading. International Journal of Innovative Research in Science, Engineering and Technology, 5(6),
9219–9227.
Rao, K.S. & Venkatesh, K.H. (1985). Uplift behavior of short piles in uniform sand. Journal of Soils
and Foundations, 25(4), 1–7.
Trochanis, A.M., Bielak, J. & Christiano, P. (1991). Three-dimensional nonlinear study of piles. Journal
of Geotechnical Engineering ASCE, 117(3), 429–447.
280

Reported correlations on compaction characteristics of fine

grained soils in the standard proctor test—a critical reappraisal
G. Sreelekshmy Pillai
N.S.S. College of Engineering, Palakkad, Kerala, India
P. Vinod
ABSTRACT: The determination of compaction characteristics is very important in the

case of earthworks construction and as such, attempts to develop empirical equations for
their prediction has been greatly received. For preliminary design and assessment, correla-
tions of compaction characteristics with index properties have been attempted by various
investigators. The reported methods for prediction of compaction characteristics of fine
grained soils in the standard Proctor compaction test are critically reviewed in this paper.
Among the various correlations, those proposed by Dokovic et al. (2013) and Vinod and
Sreelekshmy Pillai (2017), are observed to be the most accurate and precise ones. With the
use of toughness limit, which effectively takes care of the combined effect of liquid limit (wL)
and plastic limit (wP), a new set of equations for the prediction of compaction characteristics
at the standard Proctor compactive effort, are also proposed.
1 INTRODUCTION
In the construction of many earth structures, such as embankments, it is essential to determine

the compaction characteristics. The compaction characteristics of a soil, as obtained from lab-
oratory compaction tests, are the Maximum Dry Unit Weight (γd-max) and the Optimum Water
Content (OMC). The procedure for the determination of γd-max and OMC, although simple, is
time consuming and laborious since soil from highly variable sources need to be tested to assess
their suitability for their desired purpose. Delay in testing results could adversely impact project
timing and unnecessarily limit the search for sources of suitable material. Therefore, for the
preliminary assessment of the suitability of borrow materials, predictive models can be used,
especially when index properties are already known. Several attempts have been made in the
past for the determination of γd-max and OMC of fine grained soils at the standard Proctor
compactive effort. Most of them correlate compaction characteristics with the Liquid Limit
(wL) or the Plastic Limit (wP) or a combination of both. In this paper, the reported methods for
prediction of compaction characteristics of fine grained soils in terms of index properties are
critically reviewed, and a more refined correlation is proposed.
2 CORRELATIONS OF COMPACTION CHARACTERISTICS WITH INDEX

PROPERTIES AT THE STANDARD PROCTOR COMPACTIVE EFFORT
Al-Khafaji (1993) conducted studies on fine grained soils from Iraq and the USA, to find
out the relationship between Atterberg limits and compaction characteristics at the standard
Proctor compactive effort. The following equations are recommended for Iraqi soils:
OMC = 3.130 + 0.630 wP + 0.240 wL (1)

γd-max = 2.440 − 0.020 wP − 0.008 wL (2)
281

For soils of the USA, the equations are as follows:
OMC = 0.140 wL+ 0.540 wP (3)

γd-max = 2.270 − 0.019 wP− 0.003 wL (4)
An empirical method was proposed by Blotz et al. (1998) for estimating γd-max and OMC of
fine grained soils at any rational compactive effort. They used the soil data from published
literature and their experimental study. The following equations were recommended for com-
paction characteristics at the standard Proctor compactive effort:
OMC = −33.82 log wL – 0.67 wL + 43.53 (5)

γd-max = 6.29 log wL− 0.16 wL + 14.41 (6)
Gurtug and Sridharan (2004) brought out the effect of compaction energy on compaction
characteristics of fine grained soils. The analysis carried out was based on data from pub-
lished literature and experimental studies. The equations suggested were:
OMC = 0.92 wP (7)
γd-max = 0.98 γd-wP (8)
Here, γd-max was expressed in terms of γd-wP which is the Maximum Dry Unit Weight at Plas-
tic Limit Water Content, and is given by:
γd-wP = (G γw)/(1 + wPG) (9)
Studies were carried out by Sridharan and Nagaraj (2005) to determine which index prop-
erty correlated well with the compaction characteristics of fine grained soils at the standard
Proctor compactive effort. The correlation of the compaction characteristics with wP was
found to be much better than that with wL and the plasticity index. Using the data from their
study and from the literature, the following equations were recommended:
OMC = 0.92 wP (10)
γd-max = 0.23 (93.30 − wP) (11)
Based on the experimental studies on fine grained soils from different regions in Turkey,
and based on the published data, Sivrikaya (2008), also pointed out the importance of wP in
the prediction of compaction characteristics. The proposed equation for compaction charac-
teristics at the standard Proctor compactive effort being:
OMC = 0.94 wP (12)
γd-max = 21.97 − 0.27 OMC (13)
Gunaydin (2009) conducted a multiple regression analysis to determine the relationship

between percentage of soil finer than 75 µm (FC), Percentage of Sand (SC), Percentage of Gravel
(GC), Specific Gravity (G), wL, wP and compaction characteristics at the standard Proctor com-
pactive effort. The analysis was undertaken using the data of fine grained soils from Nidge,
Turkey. For prediction of OMC and γd-max, equations using wL and wP were recommended:
OMC = 0.323 wL+ 0.157 wP (14)

γd-max = 0.078 wL− 0.062 wP (15)
Dokovic et al. (2013) performed the standard Proctor tests on samples of fine grained
soils from Serbia. Multiple regression analysis was carried out to determine the relationship
between Atterberg limits and compaction characteristics, the empirical equation being:
OMC = 4.180 + 0.160 wL + 0.323 wP (16)

γd-max = 2.140 − 0.007 wL − 0.005 wP (17)
282

Varghese et al. (2013) proposed empirical equations to relate γd-max and OMC at the standard
Proctor compactive effort with soil index properties. Empirical equations were developed using
multiple regression analysis on fine grained soil data from literature. The equations were:
OMC = 5.926 + 0.613 wP − 0.026 wL (18)

γd-max = 20.159 − 0.157 wP− 0.001 wL (19)
By defining a fine grained soil by its wL and wP, the interrelationship between compaction
energy and compaction characteristics was brought within a definite framework by Sreelekshmy
Pillai and Vinod (2016), using data from literature. The recommended equations for compaction
characteristics of fine grained soils at the standard Proctor compactive effort were:
OMC = 0.172 wL+ 0.563 wP (20)

γd-max = 0.314 γd-wL + 0.742 γd-wP (21)
where γd-wL and γd-wP are Maximum Dry Unit Weight at wL and wPWater Contents, respectively.
Through a review and analysis of reported literature on compaction characteristics of fine
grained soils, Vinod and Sreelekshmy Pillai (2017) showed that Toughness Limit (wT) (which
is a function of wL and wP) bears a good correlation with γd-max and OMC; the recommended
equations for compaction characteristics at the standard Proctor compactive effort being:
Table 1. Summary of the reported empirical equations proposed by various researchers.
No. of data used

for development
Source Empirical equation of correlation Source of data
Al-Khafaji OMC = 3.130 + 0.63 wP 87 Fine grained soils of

(1993) + 0.240 wL Iraq
γd-max = 2.440 - 0.020 wP
- 0.008 wL
OMC = 0.140 wL+ 0.540 wP 15 Fine grained soils
γd-max = 2.270 - 0.019 wP fromthe USA
- 0.003 wL
Blotz et al. OMC = 34.32–33.82 logwL 22 Published data
(1998) + 0.67 wL + 9.21
γd-max 6.28 wL -2.60 + 0.16 wL + 17.02
Gurtug and OMC = 0.92 wP 86 Experimental studies
Sridharan (2004) γd-max = 0.98 γd-wP and published data
Sridharan and OMC = 0.92 wP 64 Experimental studies
Nagaraj (2005) γd-max = 0.23 (93.30 - wP) and published data
Sivrikaya et al. (2008) OMC = 0.940 wP 156 Fine grained soils from
γd-max = 21.97 - 0.276 OMC Turkey and published
data
Gunaydin (2009) OMC = 0.323 wL+ 0.157 wP 126 Fine grained soils of
γd-max = +0.078 wL + 0.062 wP Turkey
Dokovic et al. (2013) OMC = 4.180 + 0.160 wL + 0.323 wP 72 Fine grained soils of
γd-max = 2.140 - 0.007 wL- 0.005 Serbia
Varghese et al. OMC = 5.926 + 0.613 wP 95 Published data
(2013) - 0.026 wL
γd-max = 20.159 - 0.157 wP-
0.001 wL
Sreelekshmy Pillai OMC = 0.172 wL+0.563 wP 63 Published data
and Vinod (2016) γd-max = 0.314 γd-wL + 0.742 γd-wP 28
Vinod and OMC = 0.615 wT 137 Published data
Sreelekshmy γd-max = 1.134 γd-wT 102
Pillai (2017)
283

OMC = 0.615 wT (22)
γd-max = 1.134 γd-wT (23)
where,
wT = wP + 0.42 (wL− wP) (24)
and γd-wT represents the Maximum Dry Unit Weight at Toughness Limit Water Content.
Table 1 summarizes the empirical equations proposed by various researchers, source of
data used by them in the development of correlations and the number of data points used.
A critical review of the reported methods for prediction of compaction characteristics of
fine grained soils is presented in the following section.
In order to determine the accuracy and precision of the above correlations, compaction char-
acteristics of 493 fine grained soils reported in the literature review, along with their index
properties, was used. The data was obtained from the following sources: McRae (1958); Wang
and Huang (1984); Daniel and Benson (1990); Al-Khafaji (1993); Daniel and Wu (1993); Ben-
son and Trast (1995); Blotz et al. (1998); Gurtug and Sridharan (2004); Sridharan and Nagaraj
(2005); Horpibulsuk et al. (2008); Sivrikaya (2008); Gunaydin (2009); Roy et al. (2009); Patel
and Desai (2010); Datta and Chattopadhyay (2011); Beera and Ghosh (2013); Varghese et al.
(2013), Shirur and Hiremath (2014); Talukdar (2014); and Nagaraj et al. (2015).
According to Cherubini and Giasi (2000), a logical assessment of the validity of any empiri-
cal correlation can be made by an evaluation technique which simultaneously takes into con-
sideration accuracy as well as precision. Accuracy can be estimated by the Mean Value (µ)
and precision by means of Standard Deviation (s). A global evaluation of the accuracy of a
correlation can then be made by two different indices. Ranking Distance (RD), (Cherubini &
Orr, 2000) and Ranking Index (RI) (Briaud & Tucker, 1998) can be defined as follows:
   predicted value  
2
 predicted value  
RD =  1 − µ    +s 
2

 observed value    observed value   (25)
  
 predicted value   predicted value 
RI = µ ln  + s ln  (26)
 observed value   observed value 
For a good correlation, both of these indices tend to zero. The µ, s, RD and RI of the
predicted to observed values of γd-max and OMC were calculated and used to compare the
relative accuracy and precision of the above correlations (Equation 1 through Equation 8
and Equation 10 through Equation 23). A summary of the results are given in Tables 2 and 3.
It was seen that the correlations which used wT, and those that used both wL and wP as
input parameters, provided greater accuracy and precision than based on wL and wP alone.
However, as far as the prediction of γd-max is concerned, all the reported equations, except
that of Blotz et al. (1998) and Gunaydin (2009), were seen to yield satisfactory results.
More accurate and precise empirical correlations for the prediction of compaction charac-
teristics of fine grained soils was developed, taking into consideration of all the data points
hitherto reported in the literature (493 in total). The subsequent multiple linear regression
analyses resulted in the following equations:
OMC = 0.623 wT (27)
γd-max = 1.15 γd-wT (28)
284

Table 2. Comparison of the μ, s, RD and RI values of predicted to observed values
of OMC.
Source μ s RD RI
Dokovic et al. (2013) 0.99 0.19 0.19 0.22

Vinod and Sreelekshmy Pillai (2017) 1.02 0.23 0.23 0.21
Varghese et al. (2013) 0.99 0.214 0.21 0.24
Sreelekshmy Pillai and Vinod (2016) 1.06 0.23 0.25 0.25
Al-Khafaji (1993) 0.965 0.22 0.22 0.27
Blotz et al. (1998) 0.968 0.23 0.23 0.27
Sridharan and Nagaraj (2005) 1.08 0.28 0.29 0.30
Gunaydin (2009) 0.94 0.22 0.23 0.31
Gurtug and Sridharan (2004) 1.09 0.28 0.29 0.31
Sivrikaya et al. (2008) 1.11 0.28 0.30 0.32
Al-Khafaji (1993) 1.13 0.28 0.31 0.33
Table 3. Comparison of the μ, s, RD and RI values of predicted to observed values

of γd-max.
Source μ s RD RI
Dokovic et al. (2013) 1.02 0.08 0.08 0.09

Varghese et al. (2013) 1.007 0.08 0.08 0.09
Al-Khafaji (1993) 1.019 0.08 0.08 0.09
Sridharan and Nagaraj (2005) 0.99 0.08 0.08 0.09
Vinod and Sreelekshmy Pillai (2017) 0.98 0.09 0.09 0.10
Sreelekshmy Pillai and Vinod (2016) 0.98 0.09 0.09 0.10
Gurtug and Sridharan (2004) 0.985 0.08 0.08 0.104
Al-Khafaji (1993) 0.97 0.10 0.11 0.14
Sivrikaya et al. (2008) 0.99 0.13 0.13 0.15
Blotz et al. (1998) 1.08 0.20 0.22 0.25
Gunaydin (2009) 0.29 0.10 0.71 1.6
The RD and RI values of the predicted to observed values of OMC (as per Equation 27) were
0.19 and 0.18, respectively. The RD and RI values were better than that corresponding to all
reported equations in the literature review. For γd-max, RD and RI values of predicted to observed
values (as per Equation 28) were 0.09 and 0.09, respectively. These values are highly satisfactory.
4 CONCLUSIONS
A critical review and analysis of the published literature on compaction characteristics of fine
grained soils at the standard Proctor compactive effort has led to the following conclusions:
• The accuracy and precision of OMC prediction, at the standard Proctor compactive effort,
is better when wL along with wP is used.
• Among the various reported correlations, those proposed by Dokovic et al. (2013) and
Vinod and Sreelekshmy Pillai (2017) are the most accurate and precise. With the use of wT,
which effectively takes care of the combined effect of wL and wP, a new set of equations
for the prediction of compaction characteristics at the standard Proctor compactive effort
have also been proposed,and are as follows:
• OMC = 0.623 wT
• γd-max = 1.15 γd-wT
285

REFERENCES
Al-Khafaji, A.N. (1993). Estimation of soil compaction parameters by means of Atterberg limits. Quar-
terly Journal of Engineering Geology, 26, 359–368.
Benson, C.H. & Trast, J.M. (1995). Hydraulics conductivity of thirteen compacted clays. Clay Minerals,
4(6), 669–681.
Beera, A.K. & Ghosh, A. (2011). Regression model for prediction of optimum moisture content and maxi-
mum dry unit weight of fine grained soil. International Journal of Geotechnical Engineering, 5, 297–305.
Blotz, R.L., Benson, C.H. & Boutwell, G.P. (1998). Estimating optimum water content and maxi dry unit
weight for compacted clays. Journal of Geotechnical and Geoenvironmental Engineering, 124(9), 907–912.
Briaud, J.L. & Tucker, L.M. (1998). Measured and predicted axial load response of 98 piles. Journal of
Cherubini, C. & Giasi, C.I. (2000). Correlation equations for normal consolidated clays. A Nakase &
T. Tschida (Ed(s))., International coastal geotechnical engineering in practice Vol. 1, pp. 15–20. Rot-
terdam, The Netherlands: A.A Balkema.
Cherubini, C. & Orr, T.L.L. (2000). A rational procedure for comparing measured and calculated values
in geotechnics. A. Nakase & T. Tschida (Ed(s).), International coastal geotechnical engineering in
practice, Vol. 1, pp. 261–265). Rotterdam, The Netherlands: A.A Balkema.
Daniel, D.E. & Benson, C.H. (1990). Water content—density criteria for compacted soil liners. Journal
Daniel, D.E. & Wu, Y.K. (1993). Compacted clay liners and covers for arid sites. The Journal of Geotech-
nical Engineering, 119(2), 223–237.
Datta, T. & Chattopadhyay, B.C. (2011). Correlation between CBR and index properties of soil, D.K.
Sahoo, T.G.S. Kumar, B.M. Abraham & B.T. Jose (Ed(s).), Proceedings of Indian Geotechnical Con-
ference, Kochi, India (pp. 131–133).
Dokovic, K., Rakic, D. & Ljubojev, M. (2013). Estimation of soil compaction parameters based on the
Atterberg limits. Mining and Metallurgy Engineering Bor, 4, 1–7.
Gunaydin, O. (2009). Estimation of soil compaction parameters by using statistical analysis and artifi-
cial neural networks. Environmental Geology, 57, 203–215.
Gurtug, Y. & Sridharan, A. (2004). Compaction behaviour and prediction of its characteristics of fine
grained soils with particular reference to compaction energy. Soils and Foundations, 44(5), 27–36.
Horpibulsuk,S., Katkan,W. & Apichatvullop, A. (2008). An approach for assessment of compaction
curves of fine grained soils at various energies using one point test. Soils and Foundations. Japanese
Geotechnical Society, 48(1),115–126.
McRae, J.L. (1958). Index of compaction characteristics. Symposium on application of soil testing in
highway design and construction (Vol. 239, pp. 119–127). Philadelphia; ASTM STP.
Nagaraj, H.B., Reesha, B., Sravan, M.V. & Suresh, M.R. (2015). Correlation of compaction character-
istics of natural soils with modified plastic limit. Transportation Geotechnics, 2, 65–77.
Patel, R.S. & Desai, M.D. (2010). CBR predicted by index properties for alluvial soils of south Gujarat.
R. Beri (Ed.), Proceedings of Indian Geotechnical Conference, Mumbai, India (Vol. 1, pp. 79–82).
Roy, T.K., Chattopadhyay, B.C. & Ro, S.K. (2009). Prediction of CBR from compaction characteristics
of cohesive soil. Highway Research Journal, 7–88.
Shirur, N.B. & Hiremath, S.G. (2014). Establishing relationship between CBR value and physical prop-
erties of soil. Journal of Mechanical and Civil Engineering, 11, 26–30.
Sivrikaya, O. (2008). Models of compacted fine-grained soils used as mineral liner for solid waste. Envi-
ronmental Geology, 53, 1585–1595.
Sreelekshmy Pillai, G.A. & Vinod, P. (2016). Re-examination of compaction parameters of fine grained
soils. Ground Improvement, 169(3), 157–166.
Sridharan, A. & Nagaraj, H.B. (2005). Plastic limit and compaction characteristics of fine grained soils.
Ground Improvement, 9(1), 17–22.
Talukdar, D.K. (2014). A study of correlation between California Bearing Ratio (CBR) value with other
properties of soil. International Journal of Emerging Technology and Advanced Engineering, 4, 559–562.
Varghese, V.K., Babu, S.S., Bijukumar, R., Cyrus, S. & Abraham, B.M. (2013). Artificial neural net-
works: A solution to the ambiguity in prediction of engineering properties of fine grained soils.
Geotechnical and Geological Engineering, 31, 1187–1205.
Vinod, P. & Sreelekshmy Pillai, G. (2017). Toughness limit: A useful index property for prediction
of compaction parameters of fine grained soils at any rational compactive effort. Indian Geotech
Journal. 47(1), 107–114.
Wang, M.C., ASCE, M. & Huang, C.C. (1984). Soil compaction and permeability prediction models.
Journal of Environmental Engineering, 110(6), 1063–1082.
286

A study on the correlation of the shear modulus of soil with the

California bearing ratio and dynamic cone penetration value
S. Athira & S. Parvathy

College of Engineering, Trivandrum, India
V. Jaya
Department of Civil Engineering, Government Engineering College, Barton Hill, India
ABSTRACT: Stiffness of the base and sublayers is an important parameter in the design and
quality assurance of pavements. In existing pavements, prior to resurfacing, it is essential to
know the condition of the base that has been subject to traffic loading and environmental condi-
tions. When failure of a pavement occurs, a quick and accurate measurement of the properties
of the base layer is essential. The most popular methods are CBR, dynamic cone penetration
and resilient modulus tests. The bender element technique, which is applicable to a wide vari-
ety of soils, can be used on existing pavements under construction in the field. To accept a
method for design purposes, it should be validated with the conventional methods such as the
penetration methods and CBR method. This study aims at developing a correlation between the
Dynamic Penetration Index (DPI) and CBR values with the shear modulus obtained from this
method. A correlation of these values with the shear modulus can be of use to our road sector.
Keywords: shear modulus, bender element method, subgrade CBR, dynamic cone penetra-
tion testing
1 INTRODUCTION
The performance of pavements depends on the properties of the subgrade soil and pavement
materials. Accurate measurement of the properties of the base layer is essential to avoid
failure of the subgrade. Shear modulus is a soil characteristic, which determines the strength
of the subgrade and hence, its measurement is required for the design and construction of
pavements.
The shear modulus of subgrade soil can be determined in a laboratory using the theory of
wave propagation. Piezo ceramic sensors known as bender elements are currently used for the
determination of the shear modulus of soil.
A pair of in-line bender elements is usually used, where one acts as the transmitter sending
off the shear waves, while the other on the opposite end captures the arriving waves.
Dynamic Cone Penetrometer (DCP) results have never been used as an absolute indica-
tor of the in situ strength or stiffness of a material in a pavement or subgrade. The Califor-
nia Bearing Ratio (CBR) value and DPI interpretation are important soil parameters for
the design of flexible pavements and airstrips. It can also be used for determination of the
subgrade reaction of soil by using correlation. It is essential to familiarize this simple wave
propagation technique and correlate with CBR such that shear modulus can be used by prac-
ticing engineers for design and performance assessment of pavements.
This paper focuses on the measurement of the shear modulus of soil using bender ele-
ments and the separate development of a correlation with CBR and DPI.
The incorporation of bender elements in a triaxial apparatus is arguably the most common
practice for the determination of shear modulus of soil samples in the laboratory, as demon-
287

strated by Jovičić et al. (1996). The maximum shear modulus (Gmax) for saturated sand is esti-
mated from shear wave velocity measurements using bender elements. The procedure for the
selection of input parameters and the methods of signal interpretation are described by Jaya
et al. (2007). The CBR test is laborious, time consuming, depends on the skill of the technician
and the results are not accurate enough for a good quality design (Roy, Chattopadhyay & Roy,
2010). Direct evaluation of shear modulus in the laboratory and correlating it with CBR can
lead to a more simple and reliable method for CBR evaluation. This paper describes a method
of shear modulus determination using bender elements and its correlation with CBR.
Experiments were conducted in the laboratory to determine the shear modulus, DPI and
CBR value of selected subgrade soil. The properties of the materials, apparatus and method-
ology are explained in the following sections.
2.1 Materials
The soil samples used in this test were collected from selected subgrade locations in Thiruvanan-
thapuram, Kerala. The physical and engineering properties of the soil were determined accord-
ing to IS methods. The field density and specific gravity were 2.43 g/cc and 2.58, respectively.
Based on IS:1498-1970, the soil samples were found to be gravelly sand.
2.2 Methodology
The bender element test was carried out on unconfined specimens in a triaxial cell. A pair
of in-line bender elements were fixed in the test apparatus, where one acts as the transmitter
sending off the shear waves, while the other on the opposite end captures the arriving waves.
A wave generator was used to transmit sine waves into the soil samples and received signals
as waveforms on an oscilloscope. These recorded waveforms were used for interpretation of
travel time through the sample. The shear wave velocity (vs) was derived by dividing the travel
distance of the waves (between the transmitter and receiver) with the arrival time, which in
turn is squared and multiplied with the specimen’s bulk density to obtain the shear modulus.
A mathematical description of the sets of variables is the best method of scientific expla-
nation, as prior to this, in a graphical presentation there was always an element of bias or
misleading presentation (Barua & Patgiri, 1996). The study of regression enables us to get a
close functional relation between two or more variables (Kapur & Saxena, 1982).
Compaction properties were determined by a standard Proctor test as per IS:2720 (Part VII).
Unsoaked CBR values of the soil sample were determined as per the procedure laid down in
IS: 2720 (Part XVI) (1979). Dynamic cone penetration tests were conducted on the samples
according to IS:4968 (Part I) (1978).
From the results obtained, we can see that the variation of OMC with respect to CBR val-
ues was found to be inversely proportional. Variations of compaction properties with respect
to corresponding CBR values and shear modulus were determined.
From Figure 1, it may be noted that when the value of the CBR decreases there is an
increase in optimum moisture content. It was also observed that the CBR value has a signifi-
cant correlation with OMC. The relation between CBR and OMC is Y = 347.02x−1.243 with
an R2 value of 0.9503.
From Figure 2, it may be noted that the shear modulus of soil shows a similar trend
of variation with optimum moisture content. When the water content increases the shear
stiffness of the soil reduces. The relation between CBR and OMC is Y = 2.2604x-.65 with an
R2 value of 0.6708.
288

Table 1. Correlation of shear modulus with CBR value.
Sample Max dry OMC Unsoaked Shear modulus DPI

no density (g/cm3) (%) CBR Value (%) (GPa) (mm/blow)
1 1.82 15.75 7.41 0.253 12.06

2 1.65 21.25 11.11 0.216 12.26
3 1.76 15.55 11.05 0.322 13.685
4 1.76 15.55 13.96 0.252 13.86
5 1.97 13.43 10.56 0.347 11.51
6 2.05 17.6 9.71 0.347 11.01
7 1.85 17.88 12.57 0.347 13.92
8 1.95 14.9 14.57 0.346 11.13
9 1.93 12.33 12.75 0.346 9.84
10 1.93 14.81 10.93 0.346 13.54
Figure 1. Variation of CBR with OMC.
Figure 2. Variation of shear modulus with OMC.
289

Figure 3. Variation of CBR with shear modulus.
a. Correlation between CBR and shear modulus

From Figure 3 it may be noted that the CBR and shear modulus values show a trend with an
increase in OMC for all soil samples. The increase in water content reduces the stiffness of the
soil samples. Hence, its shear modulus and penetration resistance decrease with an increas-
ing water content. The failure of soil in pavements due to the presence of water is one of the
major reasons for poor performance of pavements. Evaluation of shear modulus with maxi-
mum dry density and varying water content can predict the behavior of pavements under
load with different moisture conditions. A correlation between CBR and shear modulus was
obtained as y = .0001x2 + .0143x + 1987 with R2 = 0.6056.
Based on the relation obtained between CBR and shear modulus, we can predict the value
of the shear modulus and compare it with obtained results of shear modulus.
The experimental and predicted shear modulus are almost similar for soil samples with a
CBR of less than 10%. However, for soil samples with a CBR value greater than 10%, there
is a large difference in experimental and predicted shear modulus. A comparison of shear
modulus value (Figure 4), shows that in some soil samples, the laboratory and computed
value of shear modulus are no different. The maximum difference is 7.3%, but in most cases
the minimum differences are <3%.
b. Correlation of shear modulus with DPI
The correlation equation obtained is DPI = −.026G + 20.37, where DPI = Cone penetration
index, G = shear modulus. The equation is in the form y = mx + c, which clearly implies a
relationship between the experimental variables to be linear as shown in Figure 5.
The coefficient of correlation ‘r’, obtained from the linear regression analysis is 0.94. Its
value being closer to one is a clear implication that there exists a strong functional relation-
ship between the two variables and being a negative value clearly indicates that, as the value
of small strain modulus increases, the dynamic cone penetration index decreases. Thus, that
stiff soils will have a low penetration is clearly inferred from this result.
The coefficient of determination r2, is determined to be 0.88 for the above data. This value,
which is a measure of validity of the regression model, is only 88% as other factors that
influence these two variables were not considered, thus, making the model only 88% correct.
The parity plot for the given data is shown in Figure 6, which depicts the quality of fit to
be good. As the values lie close to the parity line, normal behavior of the data is predicted.
Experimental small strain shear modulus and predicted small strain shear modulus are plot-
ted against the experimental penetration index as shown in Figure 7 and it can be seen that
only less than 10% of the predicted values do not agree with the experimentally determined
values, which clearly implies that the functional relationship between the variables is strong.
290

Figure 4. Comparison of experimental and predicted value of shear modulus.
Figure 5. Final correlation graph between small strain shear modulus and penetration index.
Figure 6. Parity plot between experimental and predicted values.
Figure 7. Comparison between experimentally determined and predicted small strain shear modulus.
291

4 CONCLUSIONS
From the literature study, it was evident that we needed to integrate the engineering behavior
of the hitherto uncorrelated tests (DCP testing, CBR method and bender element test). To
extend the application of the bender element test in the field of design of pavement subgrade,
validation with a conventional testing technique such as the DCP testing and CBR testing
was necessary. The main conclusions, highlighting the novelty of the work, are briefly men-
tioned as follows:
• The CBR values decrease with an increase in OMC and a similar trend is also observed for
shear modulus.
• The decrease of shear modulus with an increase of water content can predict the soil
strength degradation due to the presence of moisture in pavements.
• The shear modulus of the soil samples increases with a decrease in DPI after the develop-
ment of a linear regression model.
• The shear modulus of soil predicted using the derived relationship was found to be 90% in
agreement with the experimental values when the DPI was considered as the parameter.
REFERENCES
Cahyadi, J. E.C. Leong & H. Rahardjo (2009). Measuring shear and compression wave velocities of soil
using bender-extender elements. Canadian Geotechnical Testing Journal, 46, 792–812.
Dyvik, R. & Madshus, C. (1985). Proceedings of ASCE Convention on Advances in the Art of Testing
Soils under Cyclic Conditions. Detroit, Michigan, 64, 186–196.
Halsko, H.A. & Zeng, X. (2010). Piezoelectric probe for measurement of soil stiffness. International
Journal of Pavement Engineering, 11(1).
Jaya, V., Dodagoudar, G.R. & Boominathan, A. (2007). Estimation of maximum shear modulus of
sand from shear wave velocity measurements by bender elements. Indian Geotechnical Journal, 37(3),
159–173.
Lee, J.S. & Santa Marina, J.C. (2005). Bender elements: Performance and signal interpretation. Journal
of Geotechnical and Geoenvironmental Engineering, ASCE, 131, 1090–0241.
Roy, T.K., Chattapadhyay, B.C. & Roy, S.K. (2010). California bearing ratio, evaluation and estimation:
A study of comparison. Paper presented at the Indian Geotechnical Conference, Indian Institute of
Technology, Mumbai, 19–22.
Sawangsuriya, A., Bosscher, P.J. & Edil, T.B. (2005). Alternative testing techniques for modulus of
pavement bases and subgrades geotechnical applications for transportation infrastructure. ASCE,
Geotechnical Practice Publications, 3, 108–121.
292

Load-settlement behaviour of footing on laterally confined soil
P.S. Sreedhu Potty, Mayuna Jeenu, Anjana Raj, R.S. Krishna, J.B. Ralphin Rose,
T.S. Amritha Varsha, J. Jayamohan & R. Deepthi Chandran
ABSTRACT: The necessity of in-situ treatment of foundation soil to improve the bearing
capacity has increased considerably due to non-availability of good construction sites. Soil
confinement is one such method of soil improvement which can be economically adopted.
The improvement in bearing capacity and reduction in settlement of footings resting on clay
due to the addition of a laterally confined granular soil layer underneath, is investigated by
carrying out a series of finite element analyses using the FE software PLAXIS 2D. The influ-
ence of parameters like radius, depth etc. of laterally confined granular soil layer is studied.
It is observed that the load-settlement behaviour of isolated footings resting on clay can be
considerably improved by providing a laterally confined granular soil layer underneath it.
Keywords: Laterally Confined Soil, Finite Element Analyses, Load-Settlement Behaviour
1 INTRODUCTION
The decreasing availability of good construction sites has led to increased use of sites with
marginal soil properties. The necessity for in situ treatment of foundation soil to improve its
bearing capacity has increased considerably. The soil confinement is a promising technique
of improving soil capacity. This technique of soil confinement, though successfully applied
in certain areas of soil engineering, has not received much attention in foundation applica-
tions. In the last few decades, great improvements in foundation engineering have occurred,
along with the development of new and unconventional types of foundation systems through
considerations of soil-structure interaction. In the past few decades many researches have
been carried out to investigate the improvement in bearing capacity due to confining the
underlying soil. It has been proved that by confining the soil there is a reduction in the settle-
ment resulting in an increase in bearing capacity.
Much research has been carried out on soil reinforced with geosynthetics (Mahmoud
and Abdrabbo (1989), Khing et al. (1993), Puri et al. (1993), Das and Omar (1993), Dash
et al. (2001a&b), Schimizu and Inui (1990), Mandal and Manjunath (1995), Mahmoud
and Abdrabbo (1989), Rajagopal et al. (1999)). Several authors have also studied strip
foundations but reinforced with different materials (Verma and Char, 1986, Dawson and
Lee, 1988).
Sawwaf and Nazer (2005) studied the behavior of circular footing resting on confined
sand. They used confining cylinders with different heights and diameters to confine the sand.
Krishna et al. (2014) carried out laboratory model tests on square footings resting on later-
ally confined sand. Vinod et al. (2007) studied the effect of inclination of loads on footings
resting on laterally confined soil.
In this research, the beneficial effects of providing a laterally confined granular layer
underneath a footing resting on clay are investigated by carrying out nonlinear finite element
analysis using the FE software PLAXIS 2D. The influence of dimensions of the laterally
confined granular soil layer on the load-settlement behaviour, axial force distribution in the
293

confining walls, stress distribution at the interface between confining walls and surrounding
soil etc. are particularly studied.
obtained from direct shear tests; internal friction angle and cohesion intercept. Since circular
footing is used, axi-symmetric model is adopted in the analysis. The settlement of the rigid
15 noded triangular elements.
Mesh generation can be done automatically. Medium mesh size is adopted in all the simu-
lations. The size of the strip footing (B) is taken as one metre and the width and depth of soil
mass are taken as 10B in all analyses.
The footing and the confining walls are modelled using plate elements. To simulate the
interaction between the confining walls and surrounding soil, an interface element is pro-
vided on both outer and inner surfaces. The geometric model is shown in Fig. 1 and the
typical deformed shape in Fig. 2. The soil is modelled using 15-node triangular elements.
Properties of locally available sand and clay are adopted in the analyses. The material proper-
ties adopted are outlined in Table 1.
The influence of dimensions of laterally confined soil on the load-settlement behaviour and
interaction between confining walls and surrounding soil are particularly studied. Various geo-
metric parameters studied are indicated in Fig. 3. The diameter of footing is B, radius and
thickness of confined soil are r and h respectively. The diameter of footing is taken as one metre
for all the cases. Various values of r and h adopted in the analyses are indicated in Table 2.
294

Figure 2. Typical deformed shape.
Property Sand Clay
Dry Unit Weight (kN/m3) 17 16

Specific Gravity 2.3 2.1
Cohesion (kPa) 2 20
Angle of Shearing Resistance (°) 30 0
Modulus of Elasticity (kPa) 15000 7000
Poisson’s Ratio 0.25 0.3
Table 2. Values of r and h.
Parameter r/B h/B
Values 1, 1.5, 2, 2.5, 3 0.5, 1, 1.5, 2, 2.5, 3
Figure 3. Geometric parameters.
295

3.1 Load-settlement behaviour

Vertical Stress vs Settlement curves obtained from FEA, for various values of (h/B) when
r/B = are presented in Figure 4. It is seen that the load-settlement behaviour considerably
improves due to the lateral confinement of underlying soil. The improvement, when (h/B) = 3
is almost ten times when compared with that of unconfined soil at 2% settlement. The influ-
ence of radius of the confining area, when the thickness is equal to 1.5B, is presented in
Figure 5. It is seen that the load-settlement behaviour considerably improves when the radius
of confinement increases.
3.2 Interaction between confining walls and sand

The interaction between confining walls and soil in the confined area is investigated by study-
ing the distribution of normal and shear stresses at the interface. Figure 6 presents the distri-
bution of normal stresses at the interface between confining walls and sand, obtained from
FEA, for various values of (r/B) when (h/B) = 3.5.
Figure 4. Vertical Stress vs Settlement Curves for various values of (h/B); when (r/B) = 1.
Figure 5. Vertical Stress vs Settlement Curves for various values of (r/B); when (h/B) = 1.5.
296

Figure 6. Distribution of normal stress at the interface between confining walls and Sand for various
values of (r/B).
Figure 7. Distribution of shear stress at the interface between confining walls and Sand for various
values of (r/B).
It is seen that the normal stress increases with the radius of confining area upto (r/B) = 1.5
and thereafter reduces. When (r/B) = 1, the peak stress occurs at a height of B from base. For
higher values of (r/B), the point of peak stress shifts upwards and occurs at a height of 3B
from base.
Figure 7 presents the distribution of shear stress along the interface between confining
walls and sand. It is seen that the distribution of shear stress drastically changes with the
increase in radius of confining area. The pattern of soil movement within the confining area
changes with the increase in radius which influences the shear stress at interface.
It is observed that the distribution of normal and shear stresses at larger cell widths is very
different from that of smaller ell widths. This validates the observation of Vinod et al. (2007),
that at smaller cell width the confining cell—soil and footing behaves as a single unit (deep
foundation) and this behaviour changes as the radius of confinement is increased.
297

4 CONCLUSIONS
The following conclusions are drawn from the results of finite element analyses
1. The load-settlement behaviour considerably improves due to lateral confinement of under-
lying soil
2. The distribution of normal and shear stresses at the interface between confining walls and
the confined soil is considerably influenced by the radius of confinement.
REFERENCES
Binquet, J., and K.L. Lee (1975), “Bearing capacity tests on reinforced earth slabs.” Journal of Geotech-
nical Engineering Division, 101(12), 1241–1255.
Das, B.M., and M.T. Omar (1993), “The effects of foundation width on model tests for the bearing
capacity of sand with geogrid reinforcement.” Geotechnical and Geological Engineering, 12, 133–141.
Dash, S., N. Krishnaswamy and K. Rajagopal (2001 a), “Bearing capacity of strip footing supported on
geocell-reinforced sand.” Geotextile and Geomembrane, 19, 535–256.
Dash, S., K. Rajagopal and N. Krishnaswamy (2001 b), “Strip footing on geocell reinforced sand beds
with additional planar reinforcement.” Geotextile and Geomembrane, 19, 529–538.
Dawson, A. and R. Lee (1988), “Full scale foundation trials on grid reinforced clay,” Geosynthetics for
Soil Improvement. 127–147.
Khing, K.H., B.M. Das, V.K. Puri, E.E. Cook and S.C. Yen (1993), “The bearing capacity of a strip
foundation on geogrid-reinforced sand”, Geotextiles and Geomembranes, 12, 351–361.
Krishna, A., Viswanath, B. and Nikitha, K. (2014), “Performance of Square footing resting on later-
ally confined sand”, International Journal of Research in Engineering and Technology, Vol 3, Issue 6.
Mahmoud, M.A., and F.M. Abdrabbo (1989) “Bearing capacity tests on strip footing on reinforced
sand subgrade.” Canadian Geotechnical Journal, 26, 154–159.
Mandal, J.M., and V.R. Manjunath (1995), “Bearing capacity of strip footing resting on reinforced sand
subgrades.” Construction and Building Material, 9 (1), 35–38.
Puri, V.K., K.H. Khing, B.M. Das, E.E. Cook and S.C. Yen (1993), “The bearing capacity of a strip
foundation on geogrid reinforced sand.” Geotextile and Geomembrane, 12, 351–361.
Rajagopal, K., N. Krishnaswamy and G. Latha (1999), “Behavior of sand confined with single and
multiple geocells”, Geotextile and Geomembrane, 17, 171–184.
Sawwaf, M.E., and A. Nazer (2005), “Behavior of circular footing resting on confined granular soil.”
Journal of Geotechnical and Geoenvironmental Engineering, 131(3), 359–366.
Vinod, K.S., Arun, P., and Agrawal, R.K., (2007), “ Effect of Soil Confinement on Ultimate Bearing
Capacity of Square footing under eccentric-inclined load”, Electronic Journal of Geotechnical Engi-
neering, Vol. 12, Bund E.
298

Deformation behavior of sheet pile walls
H.S. Athira, V.S. Athira, Fathima Farhana, G.S. Gayathri, S. Reshma Babu,
N.P. Asha & P. Nair Radhika
Department of Civil Engineering, LBS Institute of Technology for Women, Trivandrum, India
ABSTRACT: Sheet pile walls are a common form of earth retaining structures. Earth pres-
sures developed on either side of the sheet pile wall ensure its moment and force equilibrium.
When the height of the earth that needs to be retained is rather high, the sheet pile walls are usu-
ally anchored near the top. On the other hand when the height is small, cantilever sheet pile walls
are employed. Contrary to the conventional methods, the study takes in account the stiffness and
structural capacity of sheet pile walls. The aim of the study is to analyse the deformation behav-
iour of cantilever and bulk head anchored sheet pile walls, for different depth of embedment
to height ratio by Finite Element Analysis using PLAXIS 2D software. It is observed that wall
deformation decreases when wall penetration depths are increased in cohesionless soils.
1 INTRODUCTION
Retaining walls are used to hold back soil and maintain a difference in the elevation of the
ground surface. The retaining wall can be classified according to system rigidity into either
rigid or flexible walls. A wall is considered to be rigid if it moves as a unit in rigid body and
does not experience bending deformations. Flexible walls are the retaining walls that undergo
bending deformations in addition to rigid body motion. Steel sheet pile wall is the most com-
mon example of the flexible walls because it can tolerate relatively large deformations.
Sheet pile walls are one of the oldest earth retention systems utilized in civil engineer-
ing projects. They consist of continuously interlocked pile segments embedded in soils to
resist horizontal pressures. Sheet pile walls are used for various purposes; such as large and
waterfront structures, cofferdams, cut-off walls under dams, erosion protection, stabilizing
ground slopes, excavation support system, and floodwalls. The sheet pile walls can be either
cantilever or anchored. The selection of the wall type is based on the function of the wall, the
characteristics of the foundation soils, and the proximity of the wall to existing structures.
While the cantilever walls are usually used for wall heights less than 6 m, anchored walls are
Figure 1. Sheet pile wall.
299

required for higher walls or when the lateral wall deformations are needed to be restricted.
Typically the anchors are installed when the wall height exceeds 6 m or the wall supports
heavy loads from a structure.
Contrary to conventional methods, aim of this study is to analyse the deformation behav-
iour of sheet pile walls taking into account its stiffness and structural capacity. Therefore,
this research study was conducted to understand the steel sheet pile wall behavior in terms of
deformations with different wall penetration depths.
2 NUMERICAL MODELLING AND PARAMETRIC STUDY
2.1 Numerical model

PLAXIS 2-D finite element analysis software package was used for the parametric study.
PLAXIS program is a special purpose finite element program used to perform deforma-
tion and stability analysis for various types of geotechnical applications such as excavation,
foundations, embankments and tunnels. PLAXIS, 2-D program consists of three main parts
which are Model, Calculation and Output mode. In the Model mode geometry is built. When
the geometry model is complete, the finite element model (or mesh) is generated. PLAXIS 2D
allows for a fully automatic mesh generation procedure. In the calculation mode, a number of
calculation phases will be defined.
2.2 Parametric study

The parametric study is focused primarily on studying the effect of varying wall penetration
depths larger than the ones calculated by the conventional design methods on sheet pile wall
behavior. Different types of the sheet pile walls (cantilever and anchored walls), were con-
sidered for the parametric study. The study was performed using the PLAXIS 2D version 8
Finite Element program employing 15-noded triangular elements for soil layers and 5-noded
line elements for sheet pile walls. A total of seven cases have been modeled and analyzed in
this parametric study.
The geometry of the finite element soil model adopted for the analysis is 20 × 30 m with
depths to height ratios as 0.3, 0.45, 0.6 and 0.7. The properties of soil used in the analysis are
listed in Table 1.
The material properties used in PLAXIS for the sheet pile walls are listed in Table 2
A. Cantilever sheet pile walls
The parametric study results were performed to investigate the effect of increasing wall penetra-
tion depth (D) on the cantilever sheet pile wall behavior by using medium dense sand soil (ϕ = 30°)
for 6 m high wall (H = 6 m). Figures show the models of the sheet pile wall modelled for different
D/H ratios. The total and horizontal wall displacements and also shear force for the lower and
upper ranges of the wall penetration depth are analyzed for cantilever sheet pile wall, considered
in this parametric study. The ratios adopted in this parametric study are 0.3, 0.45, 0.6 and 0.7.
Table 1. Material properties for the soil type studied.
Parameter Name Value Unit
Type of material Sand

Soil saturated unit weight γunsat 19 kN/m³
Soil unsaturated unit weight γsat 16 kN/m³
Poisson’s ratio ν 0.3
Young’s modulus E’ 1.3 × 104 kN/m²
Cohesion C’ref 1 kN/m²
Friction angle ϕ’ 30 P
Dilatency angle Ψ 0 P
300

Table 2. Material properties of the sheet pile wall.
Parameter Name Value Unit
Type of behaviour Material type Elastic

Normal stiffness EA 2.738 × 106 kN/m
Bending stiffness EI 2.3 × 104 kNm²/m
Equivalent thickness D 0.3175 M
Weight W 1.053 kNm²/m
Poisson’s ratio Ν 0.15
Figure 2. Cantilever sheet pile wall with D/H = 0.3.
B. Anchored sheet pile walls

The parametric studies were performed to investigate the effect of wall penetration depth
(D) on deformation behavior of anchored sheet pile wall by using medium dense sand soil
(ϕ = 30°) for 9 m high wall (H = 9 m). Figures show the models of the sheet pile walls mod-
elled for different D/H ratios for different anchor block dimensions. The anchor was pro-
vided at a depth of 0.6 m from the ground level.
The total and horizontal wall displacements and also shear force for the lower and upper
ranges of the wall penetration depth are analyzed for anchored sheet pile walls considered in
this parametric study. The D/H ratios adopted in this parametric study are 0.3, 0.45, 0.6 and
0.7. The anchor block dimension used is 5 × 5 cm.
3 RESULTS
A. Cantilever wall
The analysis results in terms of, maximum horizontal displacements, and shear force with
increasing wall penetration depth, for the 6 m cantilever sheet pile wall in medium dense sand
soil are given in the following figures. The wall penetration depths D, were normalized by the
wall height H, for all cases.
Wall displacements: Figures show the effect of increasing wall penetration depth on maxi-
mum horizontal wall displacements. The analyses results indicate that a significant decrease
in the wall displacements is obtained with an increase in the wall penetration depth. These
reductions are relative to the deformations of a wall designed using the conventional design
methods. The results in these figures show that by increasing the wall penetration depth to
height ratio in medium dense sand soils to 0.6 reduces the horizontal wall displacement to
about 40% of the wall displacements observed when the ratio was 0.3. There is a considerable
reduction in the wall displacements when ratio is increased to 0.7.
The graphs of D/H ratios versus total horizontal displacement were plotted for different
D/H ratios.
301

Figure 3. Anchored sheet pile wall with D/H = 0.3.
Figure 4. Horizontal displacements of can-

tilever wall. Figure 5. Shear stress distribution for dif-
ferent D/H ratios.
302

B. Anchor wall
The analysis results in terms of, maximum horizontal displacements, and shear force with
increasing wall penetration depth, for the 9 manchor sheet pile wall in medium dense sand
soil are given in the following figures. The wall penetration depths D, were normalized by the
wall height H, for all cases.
Wall displacements: Figures show the effect of increasing wall penetration depth on maxi-
mum horizontal wall displacements. The analyses results indicate that a slight decrease in the
wall displacements is obtained with an increase in the wall penetration depth. The change in
Figure 6. Deflection versus D/H for cantilever wall.
Figure 7. Total horizontal displacement for different D/H ratios for anchored wall.
303

Figure 8. Deflection versus D/H ratio.
wall displacements for all cases studied is minimal because the anchored wall is not displaced
both at the bottom of the wall and also at the anchor position. Although the wall can bend
between these positions, the overall wall displacements will be quite small with increasing the
wall penetration depth due to these fixities.
4 CONCLUSIONS
A. Cantilever walls
Analysis of a 6 m cantilever sheet pile wall in medium dense sand was done using Plaxis 2D.
As seen in the figure, increasing the wall penetration depth decreases the wall displacements.
Maximum displacement was obtained when the D/H ratio was 0.3. This indicates that a sheet
pile wall with D/H ratio 0.3 has a greater probability of failure. At the same time, minimum
displacement was obtained for D/H ratio 0.7. But, providing a D/H ratio of 0.7 is not eco-
nomical. Whereas a D/H ratio of 0.6 reduces the wall displacement by about 40%. Though
this is not as much as that obtained from D/H ratio 0.7, it is still a significant reduction.
Hence providing a D/H ratio of 0.6 is ideal.
B. Anchor walls
In the analysis of a 9 m anchored sheet pile wall in medium dense sand using Plaxis 2D, it was
found that increase in wall penetration depth does in fact that increase in wall penetration
depth does in fact reduce wall displacement but the reduction is not much. It was seen that
the displacement in each case in itself is not much, no matter the D/H ratio. This is due to the
fixity provided by the anchor. Therefore, providing a lower D/H ratio like 0.45 would suffice.
REFERENCES
Amer, Hetham A. Ramadan (2013). “Effect of wall penetration depth on the behaviour of sheet pile
walls”: A Liteature Review.
Bransby P.L.& G.W.E. Milligan (1975). “Soil Deformations near Cantilever Sheet pile walls”, Geotech-
nique 25, No. 2 175–195.
GopalMadabhushi S.P. & V. S. Chandrasekaran (2008). “Centrifuge Testing of a Sheet Pile Wall with
Clay Backfill”. Indian Geotechnical Journal, 38(1), 1–20.
Omer Bilgin, P.E. & M. Asce (2012). “Lateral Earth Pressure Coefficients for Anchored Sheet Pile
Walls”.International Journal of Geomechanics Vol 12: 584–595.
Omer Bilgin, P.E., M. Asce & M. Bahadir Erten 2009).“Analysis of Anchored Sheet Pile Wall Deforma-
tions”. International Foundation Congress and Equipment Expo.
Prakash Kumar Gupta et al., (2017). “A study relation between soil and cantilever sheet pile. A model of
theory and designing”. International journal of engineering sciences & research technology.
304

Progressive Developments in Mechanical
Engineering (PDME)
Fluid flow and fluid power systems

On the role of leading-edge tubercles in the pre-stall and post-stall

characteristics of airfoils
V.T. Gopinathan & R. Veeramanikandan

Department of Aeronautical Engineering, Hindusthan College of Engineering and Technology,
Coimbatore, India
J. Bruce Ralphin Rose

Department of Mechanical Engineering, Anna University Regional Campus, Tirunelveli, India
V. Gokul
Department of Aerospace Engineering, Madras Institute of Technology, Chennai, India
ABSTRACT: Biologically inspired designs provide different set of strategies and tools to deal
with engineering problems. The humpback whale is one such bio-inspiring species, being the
most acrobatic of the baleen whales and capable of performing good maneuvers. The presence
of large rounded protuberances or tubercles along the Leading Edge (LE) of the humpback
whale flipper highlights its uniqueness. The LE tubercles act as passive flow control devices that
improve the performance and maneuverability of the flipper. The aerodynamic performance
of NACA airfoils such as NACA 0015 and NACA 4415, and modified airfoils with leading-
edge tubercles (BUMP 0015, 4415) are numerically investigated at a Reynolds number (Re) of
1.83 × 105. The popular commercial Computational Fluid Dynamics (CFD) tool FLUENT
was used. The post-stall and pre-stall characteristics are analyzed in terms of coefficients of lift
(CL) and drag (CD) with respect to various Angles of Attack (AoAs). Both airfoils, with and
without tubercles, are investigated. Comparisons of streamline distribution, pressure coeffi-
cient (CP), CL and CD between the baseline airfoils and the airfoils with tubercles help to explain
the momentum transfer characteristics of tubercles and hence how a stall is delayed.
Keywords: Bio inspired, Humpback Whale Flipper, Tubercles, Flow separation, Stall
1 INTRODUCTION
Tubercles are rounded protuberances of the Leading Edge (LE) that alter the flow field character-
istics around an airfoil. It has been suggested that tubercles on the flipper of the humpback whale
function as lift-enhancement devices, allowing the flow to remain attached for a larger Angle of
Attack (AoA), and thus delaying stall. The protuberances found along the LE of the humpback
whale flipper vary in amplitude and wavelength across the span. Further, the amplitude of the
protuberances ranges from 2.5% to 12% of the chord and the wavelength varies from 10% to 50%
of the chord. From a more morphological point of view, Fish and Battle (1995) proved that the
geometrical properties of the tubercles could influence the aerodynamic performance of a wing.
The wind tunnel experiments of Miklosovic et al. (2004) demonstrated drastic enhance-
ments in lift for a post-stall AoA, and also showed a delay in the stall angle of up to 40%.
These experiments performed at the Reynolds number (Re) in the range of 105. One of the
mechanisms of performance enhancement is believed to be the generation of stream-wise
vortices, which improve the momentum exchange in the boundary layer. The potential ben-
efits of tubercles on the aerodynamic performance of an airfoil were addressed by Bushnell
and Moore (1991). Over the last two decades, several studies have been performed experi-
mentally and numerically to assess the influence of tubercles. The first numerical study was
307

performed by Watts and Fish (2001) using a 3-dimensional numerical panel method. They
observed a slight increase in coefficient of lift (CL) of about 4.8%, and a reduction in coef-
ficient of drag (CD) of approximately 11% in the pre-stall AoA ranges.
Johari et al. (2007) carried out a systematic experimental study on a NACA 634-021 airfoil
with tubercles of different combinations of amplitude and wavelength. The cross section
of the humpback flipper has a profile similar to the NACA 634-021. The results from these
experiments showed that incorporating LE protuberances into the airfoil generated ∼50%
higher lift without any drag penalty in the post-stall region as compared to the baseline air-
foil. While amplitude of the protuberance exhibited a notable effect on the performance of
the airfoil, it was observed that the wavelength had meager influence.
The effect of a wavy leading edge on a NACA 4415 airfoil was investigated experimentally
at a low Re of 1.2 × 105, using 2D Particle Image Velocimetry (PIV) measurement. The size of
recirculating zone and wake width was investigated experimentally by Karthikeyan et al. (2014).
A numerical investigation into the effects of a sinusoidal LE fitted to NACA 0015 and NACA
4415 profiles was analyzed with OpenFOAM software and the aerodynamic characteristics dis-
cussed by Corsini et al. (2013). Hansen et al. (2011) experimentally investigated the influence of
tubercles on NACA 634-021 and NACA 0021 airfoils. They found that for both airfoil profiles,
reducing the tubercle amplitude led to a higher maximum lift coefficient and larger stall angle.
Zhang et al. (2013) examined the effects of sinusoidal LE protuberances on two-dimensional
full-span airfoil aerodynamics at a low Re of 5.0 × 104. In terms of controlling the boundary-
layer separation, the function of protuberances is very similar in some respects to a low-profile
vortex generator. The increase in airfoil thickness causes aerodynamic deterioration at pre-stall
regimes for a wavy LE airfoil. The effect of variation in the airfoil thickness in the wavy LE phe-
nomenon was examined for NACA 0020 and NACA 0012 profiles by De Paula et al. (2016).
2 GEOMETRY
NACA 0015 and NACA 4415 profiles are used as the wing cross section. The leading-edge
tubercles design being formulated through unequal chord length. The ratio of amplitude (A)
to wavelength (λ), expressed as (η = A/λ), retains high priority in the design perspective of
tubercle research. The variable mean chord length (C), along with span-wise ordinates (z), are
elucidated by the wave equation:
C(z) = Acos(2πz /λ) + C
The chord length variation (∆c) along the span-wise direction is computed and is presented
in Figure 1.
Figure 1. Plot of chord length variation vs. span-wise ordinates.
308

Figure 2. BUMP 0015 airfoil with Figure 3. BUMP 4415 airfoil with
η = 0.05. η = 0.05.
2.1 Details of the cases analyzed

Airfoil profiles : NACA 0015 & NACA 4415
Leading-edge lines : Straight & Bump
Mean chord length : 100 mm
Amplitude : 2.5 mm
Wavelength : 50 mm
η = A/λ = 0.05
3 MESHING
A boundary is created for airfoil design in which the computational domain extends 10C
upstream and 10C downstream. The top and bottom boundaries of the domain are located
10C away from the foil. The whole computational domain is discretized with an unstructured
grid with triangular mesh, as shown in Figures 4a and 4b.
3.1 Numerical methodology

Numerical analysis is done using the FLUENT tool. The steady state compressible Reynolds-
averaged Navier–Stokes (RANS) equation was solved using a K-ω model:
 ∂U i ∂U i  ∂P ∂  ∂U i   ∂Rij 
ρ +Uk  =− + µ  +  (1)
 ∂t ∂ xk  ∂xi ∂xi  ∂x j   ∂x j 
where Rij is Reynolds stresses; U, p, ρ represents the velocity, pressure, density. The y+ values
for all the simulations are maintained below 3. Boundary conditions for numerical analysis
are pressure far field and wall.
To examine the influence of tubercles, the viscous fluid flow over a baseline airfoil is first
simulated. Then the airfoils with tubercles are simulated with the same Reynolds number.
The differing values in these simulations show the influence of the tubercles. The numerical
experiments are carried out at angles of attack of 0° to 21°. Favorable effects of tubercles are
reported on aerodynamic coefficients (CL and CD) and stalling angle.
4.1 Streamline patterns

The two-dimensional streamlines along the peak, trough and mid-span planes are determined
for all the airfoils at AoAs of 0°, 10° and 15°. For the baseline airfoil, the streamline patterns
309

Table 1. Mesh element counts.
Airfoil series Total number of elements
NACA 0015 1,323,205

NACA 4415 1,425,762
BUMP 0015 (η = 0.05) 1,136,096
BUMP 4415 (η = 0.05) 1,135,394
Figure 4. (a) Mesh generated for cambered airfoil. (b) Mesh generated for symmetrical airfoil.
are visualized along the mid-span plane. The suction surface of the baseline airfoil is in a
separated condition when the AoA is increased, as highlighted in Figures 5a, 5b and 5c, and
Figures 9a, 9b and 9c. This applies to both NACA 0015 and NACA 4415 airfoils.
The streamlines of the peak and trough regions are captured at x/b values of 0.5 and 0.25,
respectively. For the BUMP 0015 airfoil at 0° angle of attack, the separation effect is not sig-
nificant. For BUMP 0015 at an AoA of 15°, in the peak region the attached flow covers the
suction surface, forming counter-rotating vortex structures, and the flow at the trough region
is detached for the same AoA, as shown in Figures 8a and 8b. By contrast, for the BUMP
4415 airfoil at an AoA of 15°, the peak region has attached flow but the trough region has a
more separated flow than the BUMP 0015 airfoil, as highlighted in Figure 12b.
The streamline patterns over peak and trough regions at AoAs of 0º and 10º indicate that
the flow is close to the surface, as shown in Figures 6, 7, 10 and 11, with meager amounts of
separation.
4.2 Effects of tubercles on aerodynamic coefficients

The effect of aerodynamic coefficients with different angles of attack is plotted for both
baseline and modified airfoils, as shown in Figure 13. The projected plan form area are
used to calculate the CL and CD. At low angles of attack, the difference between the
baseline airfoil and modified airfoil is not significant. The NACA 0015 airfoil’s CL is
increasing and stalls at 14°. The BUMP 0015 airfoil stalls at 19°. This result indicates the
stall-delaying effect of tubercles. In particular, when considering the pre-stall operations
of the NACA 0015 airfoil, the lift coefficient is lower than for the corresponding BUMP
0015 airfoil after 10°. On the other hand, the drag coefficient difference is not significant.
The use of leading-edge bumps for cambered airfoils results in an increase in lift over
310

Figure 6. Streamlines of BUMP 0015
airfoil at α = 0°.
Figure 5. treamlines of NACA 0015 airfoil at mid-
span location.
Figure 7. Streamlines of BUMP 0015 Figure 8. Streamlines of BUMP 0015
airfoil at α = 10°. airfoil at α = 15°.
angles of attack ranging from 11° to 15°. The cambered NACA 4415 airfoil stalls at 12°
and the BUMP 4415 airfoil stalls at 15°. The drag coefficient decreased for the BUMP
4415 airfoil in the post-stall condition.
311

Figure 10. Streamlines of BUMP 4415 airfoil at
α = 0°.
Figure 9. Streamlines of NACA 4415 air-
foil at mid-span location.
Figure 11. Streamlines of BUMP 4415 air-

foil at α = 10°. Figure 12. Streamlines of BUMP 4415 air-
foil at α = 15°.
312

Figure 14. Pressure distribution plot for base-
line and modified cambered airfoils at α = 10°.
Figure 13. Comparison of lift coefficient with
angle of attack.
4.3 Effect of tubercles on CP distribution

The chord-wise pressure (CP) distributions for baseline and modified airfoils are plotted in
Figure 14. For the baseline airfoil, CP achieves a minimum value near the leading edge and,
after the peak value, CP increases along the downstream creating a strong adverse pressure
gradient. This causes the boundary layer to separate from the top surface. The most negative
values of the peak pressure developed over the suction surface of the BUMP 4415 airfoil at
α = 10° are associated with the trough cross section.
4.4 Flow field induced by tubercles

The implementation of tubercles at the leading edge of airfoil leads to the formation of
counter-rotating pairs of stream-wise vortices between the tubercle peaks (i.e. trough region).
The row of tubercles redirects the flow of air into the scalloped valley between each tubercle,
causing swirling vortices that roll up and over the airfoil, which actually enhances lift proper-
ties. The swirling vortices exchange momentum with the flow and this exchange keeps the
flow attached to the suction side of the airfoil and delays stall to higher angles of attack.
The benefits of tubercles act in a similar way to the vortex generators that are often used in
conventional aircraft. Determination of the most effective ratio of height to boundary-layer
thickness for the tubercles and optimization of tubercle geometries could serve as a potential
replacement for vortex generators.
5 CONCLUSION
The effect of leading-edge tubercles is studied on the pre-stall and post-stall behavior of an
airfoil. The formation of counter-rotating pairs of stream-wise vortices is said to enhance the
airfoil characteristics in comparison to baseline airfoils. During pre-stall, the presence of a
laminar separation bubble increases the lift of the airfoil with minimum drag effect. The impli-
cation is that tubercles on airfoils and wings helps to increase the lift beyond that of the baseline
airfoils or wings. The CL is improved from 20% to 30% for modified airfoils, as presented in
Figure 13. The bursting of the laminar separation bubble at the leading edge of the baseline
airfoil during the stall increases the drag with a significant loss in lift. In an airfoil employing
tubercles, the separation bubbles are restricted to trough regions between the tubercles. Because
the row of tubercles redirects the flow of air into the scalloped valley between each tubercle, it
causes swirling vortices that roll up and over the airfoil, which actually enhances lift properties.
Therefore, the tubercled airfoil does not stall so quickly. At post-stall angles of attack, wings
experience a reduction in lift due to flow separation. Post-stall lift can be attained in larger
amounts with the aid of tubercles with larger amplitude. The structure of the vortices increases
with the increase in amplitude of the tubercles. The stream-wise vortices convert the momen-
tum into flow, and the circulation of flow in the downstream direction increases. As a result, the
flow tends to attach to the upper surface of the wing, yielding a large increase in lift.
313

NOMENCLATURE
Re = Reynolds number

PIV = Particle Image Velocimetry
A = Amplitude
λ = Wavelength
z = Span-wise ordinate
C = Mean chord
η = A/λ
c = Chord
CL = Lift coefficient
CD = Drag coefficient
α = Angle of attack
UWS = Upper Wing Surface
LWS = Lower Wing Surface
REFERENCES
Bushnell, D.M. & Moore, K.J. (1991). Drag reduction in nature. Annual Review of Fluid Mechanics, 23,
65–79.
Cai, C., Zuo, Z., Liu, S. & Wu, Y. (2015). Numerical investigations of hydrodynamic performance of
hydrofoils with leading-edge protuberances. Advances in Mechanical Engineering, 7(7), 1–11.
Corsini, A., Delibra, G. & Sheard, A.G. (2013). On the role of leading-edge bumps in the control of stall
onset in axial fan blades. Journal of Fluids Engineering, 135, 081104.
De Paula, A.A., Padilha, B.R.M., Mattos, B.D. & Meneghini, J.R. (2016). The airfoil thickness effect
on wavy leading edge performance. Paper presented at the 54th AIAA Aerospace Sciences Meeting,
AIAA SciTech, San Diego, CA. doi:10.2514/6.2016–1306.
Edel, R.K. & Winn, H.E. (1978). Observations on underwater locomotion and flipper movement of the
humpback whale Megaptera novaeangliae. Marine Biology, 48, 279–287.
Fish, F.E. & Battle, J.M. (1995). Hydrodynamic design of the humpback whale flipper. Journal of Mor-
phology, 225, 51–60.
Hansen, K.L., Kelso, R.M. & Dally, B.B. (2011). Performance variations of leading-edge tubercles for
distinct airfoil profiles. AIAA Journal, 49(1), 185–194.
Johari, H., Henoch, C.W., Custodio, D. & Levshin, A. (2007). Effects of leading-edge protuberances on
airfoil performance. AIAA Journal, 45(11), 2634–2642.
Karthikeyan, N., Sudhakar, S. & Suriyanarayanan, P. (2014). Experimental studies on the effect of lead-
ing edge tubercles on laminar separation bubble. Paper presented at the 52nd Aerospace Sciences Meet-
ing, AIAA SciTech 2014, 13 Jan 2014, National Harbor, MD. doi:10.2514/6.2014–1279.
Lohry, M.W., Clifton, D. & Martinelli, L. (2012). Characterization and design of tubercle leading-edge
wings. Paper presented at the Seventh International Conference on Computational Fluid Dynamics
(ICCFD7–4302), Big Island, Hawaii.
Miklosovic, D.S., Murray, M.M., Howle, L.E. & Fish, F.E. (2004). Leading edge tubercles delay stall on
humpback whale flippers. Physics of Fluids, 16(5), L39–L42.
Rostamzadeh, N., Hansen, K.L., Kelso, R.M. & Dally, B.B. (2014). The formation mechanism and
impact of stream wise vortices on NACA 0021 airfoil’s performance with undulating leading edge
modification. Physics of Fluids, 26, 107101.
Skillen, A., Revell, A., Pinelli, A., Piomelli, U. & Favier, J. (2015). Flow over a wing with leading edge
undulations. AIAA Journal, 53(2), 464–472.
Watts, P. & Fish, F.E. (2001). The influence of passive leading edge tubercles on wing performance.
In Proceedings 12th UUST, Durham, NH, August 2001. Lee, NH: Autonomous Undersea Systems
Institute.
Zhang, M.M., Wang, G.F. & Xu, J.Z. (2013). Aerodynamic control of low-Reynolds number airfoil with
leading-edge protuberance. AIAA Journal, 51(8), 1960–1971.
314

Aerodynamic investigation of airfoil inspired HALE UAV
R. Saravanan, V. Vignesh, V. Venkatramasubramanian, R. Vaideeswaran &

R. Naveen Rajeev
Coimbatore, India
ABSTRACT: The present investigation focuses on structure and shape modification for a
High Altitude Long Endurance (HALE) Unmanned Aerial Vehicle (UAV). Aerodynamics and
performance parameters are defined during the conceptual design. Accordingly, the fuselage
shape is designed by NACA 2410 (low – Reynolds number) and the wing modification is car-
ried out using Gurney flaps and winglets. The objectives of this investigation are fully stated
as a consequence of the design, and the wind tunnel model is briefly described. This project
intended to get some coherent results between theoretical calculations and both the experimental
and Computational Fluid Dynamics (CFD) results. Two methods for calculating the pressure
coefficient are used, one is experimental low speed subsonic wind tunnel testing and the other
is a numerical solution from CFD ANSYS R15.0. The Spallart-Allmaras turbulence model was
used for solution initialization. The lift and drag coefficients of the airfoil at different angles of
attack were observed for both computational and experimental study. The pressure and velocity
distribution were obtained using ANSYS R15.0. The aerodynamic characteristics of the UAV
model have been carried out at different angles of attack. The aerodynamic parameters, such
as the lift coefficient, drag coefficient and (L/D) ratio are improvised by optimizing the design.
1 INTRODUCTION
An Unmanned Aerial Vehicle (UAV) is a flying machine without a human pilot on board. The
flight of UAVs as a rule work with different degrees of self-rule. Development of an advanced
unmanned aerial vehicle (UAV) for civil and military applications has driven the development of
modern aviation. In this study, a structure and shape modification procedure for a high altitude
long endurance (HALE) unmanned aerial vehicle (UAV) is presented. The improvement and
utilization of cutting edge remote detection of UAV, for example, HALE, that could more suit-
ably address the necessities of the crisis administration group, requires extensive consideration
and assessment. State-of-the-art craftsmanship in HALE stage innovation is exemplified by the
US Air Force’s Global Hawk Unmanned Endurance Air Vehicle, which is in framework trials, is
intended to go for completely independently operation (Goraj et al., 2004) with a most extreme
flight continuance of 40 hours. Worldwide Hawk could fly up to 3,000 nautical miles (5,556 km)
at up to 67,000 ft. (20, 422 m), saunter over an objective territory for 24 hours while utilizing sen-
sors or other hardware, and afterwards come back to base, all without human guidance.
The Gurney flaps (Zerihan & Zhang, 2001) are a simple device consisting of a short strip
fitted perpendicular to the pressure surface along the trailing edge of a wing. With a typical
size of 1–15% of the wing chord (Bechert et al., 2000), it can exert a significant effect on the
lift (down force), with a small change in the stalling incidence, leading to a higher CL max, as
documented by (Liebeck, 1978). Although the device was named after Gurney in the 1960s,
mechanically similar devices were employed earlier, for example, by Gruschwitz, Schrenk and
Duddy. It is anything but difficult to outline an air ship on the off chance that have infor-
mation’s about officially existing flying machines of comparative sort. It gives more fulfill-
ments and maintains a strategic distance from perplexity while picking some plan parameters
for our flying machine (Parezanovic et al., 2005). In this definite overview some numerous
315

essential outline drivers like viewpoint proportion, wing stacking, general measurements and
motor determinations are resolved for our reference. It helps with proposing another plan
and changes in our outline, which will enhance the execution of the proposed air ship. This
guarantees the execution of the flying machine according to the outline figuring’s and a sim-
ple method for planning an airplane inside specific timeframe.
2 METHODOLOGY
2.1 Computational details

Computational Fluid Dynamics (CFD) is attractive to industry as it is more financially viable
than the physical testing. In any case, one must note that the mind boggling stream reproduc-
tion is testing and blunder inclined and it takes a great deal of building ability to get approved
arrangements. Applying the basic laws of mechanics to a liquid gives the overseeing condi-
tions for a liquid. The protection of mass condition is
∂p
+ ∆. ( ρ u ) = 0
∂t
where ρ = density; u = velocity and t = time

The above conditions along with the protection of vitality condition frame an arrange-
ment of coupled and nonlinear fractional differential conditions. The design of the (Hendrick
et al., 2008) HALE UAV is undertaken using Solidworks software for the surface design. A
model was imported to an ANSYS ICEM CFD 15 in meshing tool. The model is imported
as Para solid format for getting a fine meshing a model with 160,000 elements and 260,000
nodes, the interval counts are 500,250 and 100. The model is placed in a rectangular domain.
2.2 Exploratory set-up

Four noteworthy parts of the HALE UAV display are wing, fuselage, flat stabilizer, vertical
stabilizer, winglets and Gurney folds. Mid wing sort show has been chosen for manufacture.
NACA 2410 cambered airfoil has been utilized for creation of wing, fuselage, even stabilizer
and vertical stabilizer of the said display.
2.3 Subsonic breeze burrow

Shut circuit subsonic breeze burrow has been utilized to test the manufactured HALE UAV.
The measurements of the working segment of the wind burrow are 35 cm (length) x 35 cm
(stature). Air enters the breeze burrow through an efficiently planned diffuser (cone).
A control handle could be pivoted to control the velocity from 0 to 25 m/s. The ‘L-area’
is fitted with the working segment of the breeze passage to gauge the lift, drag, weight
and their coefficients. A photograph of the subsonic breeze passage and HALE UAV
fitted with the working area of the subsonic breeze burrow are shown in Figure 3 and
Figure 4 individually.
Table 1. Boundary conditions and input parameters in fluent 14.
S. no Flow parameters Values
1 Pressure (P) in Pascal 1.0132

2 Density (ρ) in kg/m3 1.225 × 103
3 Velocity at inlet (m/s) 25
4 Reynolds number 3 × 104
316

Figure 1. Designed HALE UAV.
Figure 2. Wire frame mesh of designed HALE UAV.
Figure 4. Model fitted in a test section.

Figure 3. Low speed subsonic wind tunnels.
The lift and drag coefficients of the airfoil at different angles of attack were observed for
both computational and experimental study. Data is taken by varying the angle of attack and
by keeping the free stream velocity constant for both tests.
3.1 Comparison between computational and experimental data

Experiment results are compared with computational results and it illustrates the similarities
and minimum contrast behavior between computational and experimental data. Results are
calculated by keeping free stream velocity as a constant and varying the angle of attack. The
plot is drawn between locations of pressure tapings and pressure coefficient. The plots shown
will give the pressure coefficient at various angle of attack for both numerical and analytical
results. Figure 5, the pressure at the suction region decreases because of decreasing the lift
coefficient value and this scenario is similar for both experimental and numerical calculations
and are become evident through the plot. In Figure 7 the pressure at suction zone tends to
increase by simultaneously increasing the lift coefficient. Both CFD and experimental results
indicate the similar traits, thus in turn to improve the aerodynamic parameters (Jahangir Alma
et al., 2013). In Figure 10, the pressure difference value reaches a peak point and the lift coef-
ficient value drops rapidly such that the stall region is approached in both experimental and
CFD results. The computational results for the HALE UAV are obtained by using the Sparlart
Allmaras turbulence model. Keeping the inlet velocity at 25 m/s and by varying the angle of
attack, the velocity distribution and pressure distribution is obtained as contour diagrams.
Thus the values of the lift coefficient, drag coefficient and pressure coefficient is obtained.
The value of pressure coefficient is drawn in the graph against the location of pressure
tapings, where the location is plotted by using the solver. The lift coefficient plotted with
angle of attack gives the lift curve and the drag coefficient is plotted against angle of attack
317

Figure 6. Comparison of pressure coefficient
vs. locations between computational and experi-
vs. locations between computational and experi-
mental data at α = 8° @ V = 25 m/s.

Figure 7. Comparison of pressure coefficient vs. locations between computational and experi-
vs. locations between computational and experimental data at α = 16° @ V = 25 m/s.

vs, locations between computational and exper- Figure 10. Comparison of pressure coefficient
imental data at α = 24° @ V = 25 m/s. vs. locations between computational and experi-
to visualize its effect at various levels. The drag variation is represented in the polar diagram
in which CL is plotted against CD. In this plot, lift coefficient increases until a particular angle
of attack after which it drops due to subsequent increase in the drag coefficient. The graph
between pressure coefficient and location of pressure tapings is plotted for both experimental
as well as computational results. The obtained results are compared and validated. It is found
that the optimized lift coefficient and the stall region of the designed HALE UAV is similar
in both experimental and CFD results.
318

4 CONCLUSIONS
The primary goal of this paper is to investigate the aerodynamic characteristics of HALE
unmanned aerial vehicle (UAV) in order to make a UAV with optimized aerodynamic per-
formances. To that end, it explains how improvisation of aerodynamic parameters such as
coefficient of lift (CL) and coefficient of drag (CD) is done. The coefficient of pressure (CP)
for the model is calculated from the experimental results. The report concludes the result by
using the methodology that has been implemented in it. The desired result is achieved by
modifying the shape of the fuselage. The presence of a Gurney flap on the wing served to
increase the lift generated on the body. The presence of winglets makes the flow laminar by
preventing the creation of vortices. The induced drag has been counterattacked by using the
winglets and the vortex drag has been counterattacked by Gurney flaps. For three dimen-
sional analysis, favorable results were obtained with the ANSYS, FLUENT software. The CL
Vs. α and CD Vs. α graphs give the satisfactory results with values of CL and CD. Different
data is obtained by keeping the free stream velocity constant and by varying the angle of
attack for both the experimental and investigational study. Since HALE UAVs fly at high
altitude, and in low density conditions at lower speeds, it is difficult to locate the transition
point. This problem is overcome by using a low Reynolds number airfoil shaped fuselage
structure. Manometric readings are obtained using a subsonic wind tunnel at a constant
velocity V = 25 m/s. From the manometric readings, free stream static pressure and total
pressures were calculated. These manometric readings are calculated with the aid of pressure
tapings. Computational investigations have been performed to examine the effectiveness of
the aerodynamic parameters such as CL, CD and CP for the HALE UAV. The stall angle for
the designed HALE UAV is 28°. The coefficient of lift is found to be maximum at the angle
of attack α = 16°. Experimental results compared with computational simulations are next
in relation to the aerodynamic coefficients. The pressure contour at stall angle of attack is
found to be maximum at the pressure side and minimum at the suction side. The velocity
contour and pressure distribution exhibits the same result. Finally, the result concludes that
the aerodynamic parameter CP has been efficient for experimental and computational results.
This project achieved some coherent results between theoretical calculations and both the
experimental and CFD results. This analysis yields better results for a HALE UAV and can
be implemented on future UAV projects to get the optimized coefficient of lift (CL) value.
Using this design optimization, the HALE UAVs will be able to achieve higher ranges and
longer endurance.
REFERENCES
Bechert, D.W., Meyer, R. & Hage, W. (2000). Drag reduction of air-foils with mini flaps—AIAA. Berlin,
Germany.
Goraj, Z., Frydrychewicz, A., Ašwitkiewicz, R., Hernik, B., Gadomski, J., Goetzendorf-Grabowski,
T., Figat, M. & Chajec, W. (2004). High altitude long endurance unmanned aerial vehicle of a new
generation-a design challenge for a low cost, reliable and high performance aircraft. Bulletin of the
Polish Academy of Sciences Technical Sciences, 52(3), 173–194.
Hendrick, P., Verstraete, D. & Coatanea, M. (2008). Preliminary design of a joined wing HALE UAV.
26th International Congress of the Aeronautical Sciences, (ICAS2008).
Jahangir Alam, G.M., Md. Mamun, Md Abu, T.A, Md. Quamrul Islam, Md. & Sadrul Islam, A.K.M.
(2013). Investigation of the aerodynamic characteristics of an airfoil shaped fuselage UAV model,
International Conference on Mechanical Engineering.
Liebeck, R.H. (1978) Design of subsonic air-foils for high lift. Journal of Aircraft, 15(9), 547–561.
Parezanovic, V., Rasuo, B. & Adzic, M. (2005). Design Airfoils for wind turbine blades. Research
Gate Publications. Retrieved from https://www.researchgate.net/publication/228608628_
DESIGN_OF_AIRFOILS_FOR_WIND_TURBINE_BLADES.
Zerihan, J., & Zhang, X. (2001). Aerodynamics of Gurney flaps on a wing inground effect, AIAA
JOURNAL, 39(5), 772–780.
319

Industrial engineering and management

The usability of road traffic signboards in Kottayam
R. Rajesh, D.R. Gowri & N. Suhana

ABSTRACT: Traffic signboards provide important information, directions and warnings on

the road; they are designed and placed to provide assistance to drivers. In this study, the sign-
board usability issues of drivers in Kottayam district was studied. The significant factors that
affect usability are determined from a literature survey, and a preliminary observational field
study was conducted. Next, based on the brainstorming sessions, a cause and effect diagram
for the poor usability of signboards was developed. A questionnaire survey was conducted. A
majority of the driver respondents (58%) do not use the signboards. The reasons for non-usage
or poor usage are attributed to operational factors, environmental factors, or human factors.
The respondents perceived that poor signs would have the largest negative impacts on driving
control and confusion. Numerous factors relating to driver, road, signboard, and environment
affect the usability of traffic signs. Signboard factors such as readability, font size, language,
position, long-distance visibility, and multi-sign configuration, road factors such as advertise-
ments, greenery, buildings, and speed, driver factors such as driver age, eyesight, road familiarity,
and driver experience, and environmental factors such as weather, light, and police presence, all
affect traffic sign usability. Traffic signs could be made better by applying ergonomic principles
and a number of suggestions to improve traffic sign usability are provided. The study is limited
by its sample size and a larger sample is needed to draw a generalized conclusion. A field-based
observational study or a naturalistic driver study could provide for more quantitative analysis.
Keywords: Traffic signboards, warnings on the road, multi-sign configuration
1 INTRODUCTION
Traffic signboards provide important information, directions and warnings on the road; they
are designed and placed as assistance to drivers. They keep traffic flowing freely by helping
drivers reach their destinations and letting them know entry, exit, and turn points in advance.
Pre-informed drivers will naturally avoid committing mistakes or taking abrupt turns and
causing bottlenecks. Comprehension of traffic signboards is crucial to safety, but they are not
always detected or recognized correctly. Signboards present issues in terms of detection and
recognition due to poor visibility, bad weather conditions, the color combinations used, their
height and position, vehicle speed, and driver’s age and vision.
Usability indicates ease or convenience of use. In this study, we consider signboard usabil-
ity issues. The objective of this paper is to assess the usability problems of signboards faced
by drivers and to study the effectiveness of signboards in Kottayam.
2 LITERATURE REVIEW
Numerous studies have indicated that regulatory and warning signs help to improve the flow
of traffic, reduce accidents and ensure that pedestrians can safely use designated crosswalks.
Traffic sign usability is influenced by numerous factors.
323

It is essential for increased road safety that drivers can understand signs correctly and
quickly (Cristea & Delhomme, 2015). Meaningfulness, recognition, simplicity, and symbol
design affect sign comprehension (Yuan et al., 2014). Familiarity, standardization, simplicity,
and the symbol-concept compatibility of traffic signs also influence comprehension (Ou & Liu,
2012; Shinar & Vogelzang, 2013). For better usability, Jamson and Mrozek (2017) emphasized
color and shape, while Khalilikhah and Heaslip (2016) stressed sign condition. Traffic signs,
traffic factors, and driver factors affect the driver and their safety performance in relation to
accuracy, correctness of answers, and reaction time (Ng & Chan, 2008; Di Stasi et al., 2012;
Shinar & Vogelzang, 2013; Yuan et al., 2014; Kazemi et al., 2016; Domenichini et al., 2017).
A study of signboard comprehension by Ben-Bassat and Shinar (2015) indicated that younger
drivers performed significantly better than older drivers on both accuracy and response time.
3 METHODOLOGY
The significant factors that affect usability have been determined from the literature survey. A
preliminary observational field study was carried out on three 15-kilometer stretches of road
in the Kottayam district of Kerala state in India. The field study provided an opportunity
to comprehend the signboard types and the factors highlighted in the literature that affect
their usability. Figures 1 and 2 show some of the signboards on the roads observed. Subse-
quently, a brainstorming session was undertaken with four student project members and two
experts to focus on the road signboards’ attributes and their usability. Figure 3 shows a cause
and effect diagram for the poor usability of signboards according to these brainstorming
sessions. Next, a four-part questionnaire was prepared and a pilot survey was conducted in
Kottayam district. Convenience sampling was used for the survey. Where possible, respond-
ents were approached individually using hard copy; others were surveyed using an online
method. A total of 236 responses were obtained. These were filtered for missing values and,
subsequently, descriptive statistics were used for data analysis.
The cause and effect diagram (Figure 3) highlights four components that affect usability,
that is, driver, signboard, road, and environment. Driver factors include age, gender, experi-
ence, familiarity with signs, familiarity with roads, and driver behavior; road factors include
Figure 1. Signboards with two languages, color combination, variable font size, and graphics.
Figure 2. Damaged, vegetation-obstructed, and poor-visibility signboards.
324

Figure 3. Cause and effect diagram for signboard usability.
speed of vehicle, type of road, traffic conditions, road infrastructure, and advertisements;
signboard design attributes include shape, size, color, content, position, and condition; envi-
ronmental factors include weather, light, and monitoring system. Numerous studies have
provided evidence of the differing impacts of various factors on traffic sign effectiveness.
For example, sign layout, shape, and familiarity would improve driver comprehension and
hence driver response, while damaged signs, obstructing vegetation, and advertisements
would make signs unusable for the driver. Higher speed, driver age, and traffic volumes would
impair sign usability as a result of the shorter driver response times necessitated by them.
4 RESULTS
Kottayam contains 406 km of state highways and a total road length of 3,449 km. The three
main types of signs, that is, mandatory signs, cautionary signs, and information signs, have
been used regularly along the roads (see, for example, Figures 1 and 2). The languages used
are predominantly Malayalam and English. The color combination for the text on informa-
tion signs is white text on a green background.
The sample was relatively young (mean = 25.19 years; SD = 8.863 years) with 91% male
respondents. A proportion (50%) of respondents used two-wheelers. A majority of drivers
(58%) reported that they do not use signboards or only use them sometimes (36%). The
reasons for non-usage or low usage could be attributed to operational factors, environmental
factors or human factors. Some of these aspects are examined in the following sections.
4.1 Impact of poor traffic signboard usability

Poor and improper use of signboards causes numerous issues to drivers, for example, confu-
sion, traffic congestion, time loss, penalties, loss of driving control, and accidents. In this
survey, it has been found that poor signboards lead to traffic congestion (50%), time loss
(24%) and confusion (13%). Two probable reasons for non-usage or poor usage of signs
are: (i) drivers are familiar with the roads and don’t perceive a need to use the signboards;
(ii) drivers have difficulty in using the signboards. The two major reasons for poor signboard
usability are damaged signage (49%) and fading colors (33%). Interestingly, ‘sign knowledge’,
‘user behavior’ and ‘time pressure’ have not been indicated by the respondents as the primary
reason for poor signboard use.
325

Among the six impacts assessed, that is, ‘driving control’, ‘confusion’, ‘time loss’, ‘traf-
fic congestion’, ‘incidents’, and ‘traffic compliance’, respondents indicated that poor traffic
signs cause the most negative impact on ‘driving control’ (88%) and ‘confusion’ (83%); 68%
of respondents felt that poor traffic signs cause ‘time loss’, 66% that they would lead to
‘traffic congestion’, and 51% indicated they would lead to ‘incidents’. In terms of ‘traffic
compliance’, respondents were divided: 49% indicated that poor traffic signs would have
a negative impact on this, while 31% indicated there would be no effect. A statistical T-test
found significantly greater negative effects from poor traffic signs on ‘congestion’, ‘time loss’,
and ‘driving control’ for those with less than five years’ driving experience.
4.2 Factors affecting signboard usability

Figure 4 shows the signboard attribute rating. Shape attributes (color combination, height,
size, layout) have been favorably rated, while readability (inside text, font size, language) and
visibility (weather, distance, night-time, position and configuration) attributes have been nega-
tively rated. Readability and visibility attributes are among the likely causes of poor signboard
usability. In addition, vehicular (speed), environmental (greenery, buildings, advertisements)
and human (age and eyesight) factors produce a further negative effect on usability. As far as
other human factors (driving experience and road familiarity) are concerned, they are per-
ceived to produce a positive effect on usability. This implies that there could be a positive
trade-off in the attitude to compliance of drivers regarding road traffic signboard use.
There are numerous external factors affecting traffic signboard usability, such as high vehi-
cle speed, greenery and vegetation, presence of buildings, presence of flags/advertisements/
placards, increase in age of drivers, driver’s eyesight, driver’s compliance attitude, driver’s
familiarity with road, driver’s experience, and police presence (see Figure 5). Among these,
high vehicle speed, greenery and vegetation, presence of buildings, presence of flags/adver-
tisements/placards, increase in age of drivers, and driver’s poor eyesight have a negative effect
on signboard usability, while driver’s familiarity with the road, driver’s experience, and police
presence have a positive effect on signboard usability. During the field study, it was observed
that the roads had significant bends, were densely populated with buildings, and had signifi-
cant vegetation along them. Some of this is visible in Figures 1 and 2. Details of the field
study would be made available in the part-2 of this article. Contrary to the expectation that
a driver’s attitude to compliance might have a positive effect on signboard usability, it was
found that only 27% of respondents perceived that compliance attitude had a positive effect
on signboard usability. This is a behavior issue that needs to be further examined.
Figure 4. Signboard attribute rating.
326

Figure 5. Effects of road conditions and human factors on road signboard usage.
4.3 Discussion
Numerous factors relating to driver, road, signboard, and environment affect the use of traf-
fic signs (see Figure 3). Readability, font size, language, position, long-distance visibility, and
multi-sign configuration are prominent attributes that affect sign usability. Advertisements,
greenery, buildings, and high speed are road factors that reduce traffic sign usability. Driver
factors such as driver age and eyesight are perceived to have a negative effect, but road famili-
arity and driver experience are perceived to have a positive effect on traffic sign usability. The
environmental factors of weather, light, and police presence affect traffic sign usability. For
better use and compliance with road signage there is a need to redesign traffic signs. A user-
centered design of traffic signs that considers driver, road, signboard, and environmental fac-
tors would enable better comprehension of the signs, their more effective usability, and better
driver responses. Ultimately, this would lead to better traffic flow and safety performance.
Traffic signs could be made better by using ergonomic principles (Jamson & Mrozek, 2017).
From the survey carried out, 32% of the respondents felt that an on-board recognition
system could improve sign usability. A good traffic sign could help in better design of intelli-
gent driving-aid systems (Amditis et al., 2010; Wen et al., 2016). The respondents emphasized
the use of reflective signs (30%) and improved road infrastructure (20%) for enhanced traffic
sign usability.
We make the following specific suggestions:
– Signboards should be cleaned and maintained every six months.
– The font size, line spacing and sign size should be as per the Indian standards of motor
vehicle legislation.
– Focus on better content layout in signs for better readability.
– Position and align signboards to maximize long-distance visibility.
– Remove greenery regularly from the lines of sight of signs.
– Provide for night-time sign visibility through reflective signs or night lighting.
– Examine the possibility of providing user-compatible in-vehicle traffic sign devices that
help to address human age or eyesight issues.
– Increase monitoring efforts for traffic sign compliance.
5 CONCLUSION
The road traffic sign usability issues of drivers in Kottayam district were studied. A majority
of the drivers surveyed (58%) do not use the signboards or only use them sometimes (36%).
327

The reasons for non-usage or poor usage are attributable to operational factors, environmen-
tal factors or human factors. The two major issues that the respondents have experienced
due to poor signs are traffic congestion and time loss. The respondents perceived that poor
signs have the largest negative impact on driving control and confusion. Numerous factors
relating to the driver, road, signboard, and environment affect the usability of traffic signs.
Signboard factors such as readability, font size, language, position, long-distance visibility,
and multi-sign configuration, road factors such as advertisements, greenery, buildings, and
speed, driver factors such as driver age, eyesight, road familiarity, and driver experience, and
environment factors such as weather, light, and police presence, all affect traffic sign usability.
Traffic signs could be made better by applying ergonomic principles. A number of suggestions
to improve traffic sign usability have been identified. These include signboard cleaning and main-
tenance, use of Indian traffic standards, better content layout design, positioning of signboards
for maximum visibility, removal of obstructions and greenery along lines of sight, improved
reflectivity, the provision of in-vehicle traffic sign devices, and increased monitoring efforts.
This study was limited by its sample size and a larger sample is needed to allow a more
generalized conclusion. A field-based observational study or a naturalistic driver study could
provide for more quantitative analysis.
ACKNOWLEDGMENTS
We thank Mrs. Asha Lakshmi, Ms. Harsha Surendranand, Dr. Vinay V. Panikar, Dr. Basha
S.A, and students of M. Tech 2016 for helping us carry out this study.
REFERENCES
Amditis, A., Pagle, K., Joshi, S. & Bekiaris, E. (2010). Driver–Vehicle–environment monitoring
for on-board driver support systems: Lessons learned from design and implementation. Applied
Ergonomics, 41(2), 225–235.
Ben-Bassat, T. & Shinar, D. (2015). The effect of context and drivers’ age on highway traffic signs com-
prehension. Transportation Research Part F: Traffic Psychology and Behaviour, 33, 117–127.
Cristea, M. & Delhomme, P. (2015). Factors influencing drivers’ reading and comprehension of on-
board traffic messages. European Review of Applied Psychology, 65(5), 211–219.
Di Stasi, L.L., Megías, A., Cándido, A., Maldonado, A. & Catena, A. (2012). Congruent visual infor-
mation improves traffic signage. Transportation Research Part F: Traffic Psychology and Behaviour,
15(4), 438–444.
Domenichini, L., La Torre, F., Branzi, V. & Nocentini, A. (2017). Speed behaviour in work zone crosso-
vers. A driving simulator study. Accident Analysis & Prevention, 98, 10–24.
Jamson, S. & Mrozek, M. (2017). Is three the magic number? The role of ergonomic principles in cross
country comprehension of road traffic signs. Ergonomics, 60(7), 1024–1031.
Kazemi, M., Rahimi, A.M. & Roshankhah, S. (2016). Impact assessment of effective parameters on
drivers’ attention level to urban traffic signs. Journal of the Institution of Engineers (India): Series
A, 97(1), 63–69.
Khalilikhah, M. & Heaslip, K. (2016). The effects of damage on sign visibility: An assist in traffic sign
replacement. Journal of Traffic and Transportation Engineering, 3(6), 571–581.
Ng, A.W. & Chan, A.H. (2008). The effects of driver factors and sign design features on the comprehen-
sibility of traffic signs. Journal of Safety Research, 39(3), 321–328.
Ou, Y.K. & Liu, Y.C. (2012). Effects of sign design features and training on comprehension of traffic signs
in Taiwanese and Vietnamese user groups. International Journal of Industrial Ergonomics, 42(1), 1–7.
Shinar, D. & Vogelzang, M. (2013). Comprehension of traffic signs with symbolic versus text displays.
Traffic & Transportation Research, 44(1), 3–11.
Wen, C., Li, J., Luo, H., Yu, Y., Cai, Z., Wang, H. & Wang, C. (2016). Spatial-related traffic sign inspec-
tion for inventory purposes using mobile laser scanning data. IEEE Transactions on Intelligent
Transportation Systems, 17(1), 27–37.
Yuan, L., Ma, Y.F., Lei, Z.Y. & Xu, P. (2014). Driver’s comprehension and improvement of warning
signs. Advances in Mechanical Engineering, 6, 582–606.
328

Development and analysis of robust neighbourhood search for

flow-shop scheduling problems with sequence dependent setup times
V. Jayakumar & Rajesh Vanchipura

Department of Mechanical Engineering, Government Engineering College, Thrissur, Kerala, India
ABSTRACT: This paper focuses on the problem of determining a permutation sched-

ule for n jobs in an m-machine flow shop environment. It operates in a Sequence Depend-
ent Setup Time (SDST) environment. The objective function considered is minimization
of makespan. A new heuristic algorithm called, ‘Robust Neighbourhood Search’—a local
search algorithm with added global search properties is used for minimization of makespan,
the objective. The proposed heuristic involves an explorative search with random variables
converge to near optimal solution. More than one cycle is used to ensure achieving results
close to global minima. For purpose of experimentation, 1080 SDST benchmark problem
instances are used. That involves nine problem sizes at eight different levels of setup times.
Graphical analysis, relative performance index analysis, and statistical analysis are carried
out on the makespan results obtained for all the benchmark problem instances. The analysis
reveals that for all problem instances the RNS works superior to the Genetic Algorithm.
1 INTRODUCTION
The production industries generally involve various processes such as design, manufacturing,
marketing. And the manufacturing process consists of operations like, welding, milling etc.
In reality there can be numerous of operations to be carried out on a raw material before it
comes out as a final product. Many times other materials were added to or removed from
the original material, some totally different in nature. Each of this operations consume dif-
ferent processing times. In addition to that, each of these operations need separate machines
to perform each action. Hence, in the shop floor each machine is used for a particular job
at a time. Once a job is done, the setup on the machine needs to be changed to do another
job. There will be some time required to change the setup done for doing one job to another
job. The setup times are invariably involved in all scheduling situations. However, they are
added to processing times in many of the situations. Certainly, this procedure will reduce the
complexity of the problem solved. On the other hand it will affect the quality of solutions
obtained. Hence, there is need for explicit consideration setup time, which is addressed in the
present study.
In shop floor there are many configurations such as single machine, parallel machine, flow
shop, job shop etc. However, real manufacturing situations encounter numerous variations
of these basic shop configurations. It is observed that more than one-third of production sys-
tems follow flow shop configuration (Foote and Murty, 2009). A flow shop is characterised
by the flow of work that is unidirectional i.e., there are n jobs to be processed on m machines.
The order jobs are processed on machines is assumed to be same i.e., permutation flow shop.
If there are n jobs, there are n! total number of solutions. The present research considers a
realistic variation of the general flow shop, i.e. a flow shop operating in a sequence depend-
ent setup time environment. When the setup time is added, the complexity of the problem
becomes NP complete in nature (Jatinder & Gupta, 1986).
The flow shop scheduling problems are widely used in industry. Reducing the makespan
is the main objective that is desired in most of the situations. Exact solution methodologies
329

such as branch-and-bound, dynamic programming explored by researchers but are limited
small size problems involving two or three machines. Since there are no solution methodolo-
gies giving exact solution for these problems, heuristic methods are widely employed. The
heuristics can be constructive or improvement in nature. Constructive heuristic gives a single
solution all the time while improvement heuristics give different solutions every time. The
advantage is that the algorithms can be easily changed to meet some specific requirements.
Both have their own merits and demerits.
Improvement heuristics are applied more than constructive ones owing to their flexibil-
ity. Majority of the improvement heuristics are meta-heuristics such as genetic algorithm,
particle swarm optimization algorithm, and simulated annealing algorithms etc. By observ-
ing the nature or otherwise, scientists and industrialists try to make new heuristics to solve
flow shop scheduling problems. Such an experiment is what we try to do in this paper. The
present paper proposes a novel variation of neighborhood improvement heuristic for Flow
shop scheduling problems with setup times.
2 LITERATURE REVIEW
The flow shop scheduling problems have been an intense subject of study over past few dec-
ades. The literature review carried out in the present paper focuses on the works done on flow
shop scheduling with and without set up times for jobs. The earlier works are done without
taking the setup times into consideration. Both constructive and improvement heuristics are
worked on it quite well. It can be seen that make span minimization is given the premium
importance in all of them. For practical size problems, the researchers have used either con-
structive or improvement heuristics.
Hence the literature review can be divided into two sections.
2.1 Constructive heuristics

For the SDST flow shop scheduling problems, the work of Rios-Mecardo and Bard (1998)
stand a good constructive algorithm. They present a constructive algorithm known as
NEHRB, which is an extension of another well known constructive algorithm of Nawas
Enscore Ham. The same NEH is again modified by adopting idea of job insertion to it by
Chakraborty & Laha (2009). Another significant work done is by Vanchipura & Sridharan
(2012) where the idea of fictious jobs which are pair of jobs with highest processing time
taken as one job and added into the sequence in decreasing order.
2.2 Improvement heuristics

Rios-Mecardo and Bard (1998) has also given an improvement heuristic called the greedy
randomised adaptive search procedure (GRASP). Das et al. considers problem instances
while setup times are quite higher than the processing time and focuses on saving time.
The well known meta heuristic Simulated Annealing is worked on by Parthasarathy &
Rajendran (1997). Ruiz et al. (2005) does the GA analysis for hybrid flow shop with
sequence dependent set up times with four new cross over operators. Bat intelligence
approach was developed for problems without setup times by Malakooti et al. (2007). Liu
(2007) does the Particle swarm optimisation technique. In particular, the PSO applies the
evolutionary searching mechanism which is characterized by individual improvement,
population cooperation, and competition to effectively perform exploration. Marichel-
van (2002) does an improved hybrid Cuckoo Search metaheuristic. He found that the
results are good but as the problem size increased, it was necessary to change search
characteristics. Yagmahan & Yenisey (2007) do the Ant-colony optimization method.
They study the multi objectives of makespan, total flow time and total machine idle time.
Taillard (1989) gives benchmark problems for permutation flow shop, job shop and open
shop scheduling problems.
330

3 PROBLEM FORMULATION
3.1 Experimental problem description

There are altogether 1080 problem instances which are tested with the new heuristic method
developed. For example one sample problem can be a 20 job problem with 5 machines having
setup times of job equal to 0% of processing time. These problems are made from Taillard
Benchmark problems.
For example one sample problem can be a 20 job problem with 5 machines having setup
times of job equal to 0% of processing time. These problems are made from Taillard Bench-
mark problems.
There are 120 problem instances of Taillard are available for general flow shop. But these
have only processing time only. To add setup times also to it, the setup time matrix is so
formed at nine levels as percentages of processing time as described above.
4 SOLUTION METHODOLOGY
The present study proposes a novel variation of neighbourhood search for scheduling SDST
flow shop. Generally, all neighbourhood search procedures use only one type of neighbour-
hood. In order to provide more intensity to the search procedure two different types of neigh-
bourhood are searched in the same local search. These are swap and insertion neighbourhoods.
In the swap neighbourhood, two job numbers are randomly generated and their positions are
mutually exchanged. In the insertion neighbourhood procedure, a job number is randomly
generated. The randomly generated job is inserted in the random position generated. The
advantage of this procedure is that both these neighbourhoods generate mutually exclusive
set of solution sequences, which result in intensified search. The proposed heuristic method
involves local and global search parameters which are optimized to make it a robust algorithm.
The working procedure of the algorithm is as shown.
The local search
331

The global search
The makespan results are obtained for all the 1080 problem instances. They are compared
with results of Genetic algorithm obtained from Vanchipura R and Sridharan R. The vari-
ous analysis methods used are Graphical Analysis, Relative performance index analysis and
Statistical analysis. It is found that the Robust Neighbourhood search performs superior to
GA for all problem instances and as the problem size increases in terms of both setup times
and number of jobs, the difference become more and more evident.
5.1 Graphical analysis

The RNS is tested for altogether 1080 problem instances. They are grouped into 9 groups
based on set up times. In each group, there are twelve different problem instances. Each prob-
lem instance has 10 sample problems. Makespan values obtained using RNS are compared
with that of GA. Setup times taken are 1%, 5%, 10%, 25%, 50%, 75%, 100%, 125%, 150%.
5.2 The relative performance index analysis

The relative performance of these two algorithm is the difference of Robust Neighbourhood
Search based on the Genetic Algorithm. It can be mathematically stated that
RPI = (RNS – GA/GA)
For the two sets of 200 jobs problems, ie the 200*10 and 200*20 the RPI values are respec-
tively −0.02706 and −0.044. The RPI value for 500*20 problem is −0.06411. These results
clearly shows that RNS performs better than GA.
Out of this 100% setup time level is taken as the base case and graphs are plotted. There
are 12 graphs corresponding to 12 different size problems at 100%. Due to space limitations,
5 graphs are shown (Figs. 1 to 2) corresponding to 20*10, 50*10, 100*10, 200*10 and 500*20
size problems.
Figure 1. SDST 100, 20 Jobs & 10 Machines.
332

Figure 2. SDST 100, 50 Jobs & 10 Machines.
The result clearly shows that the new algorithm, ie the Robust Neighbourhood Search
performs better than GA for all problem instances.
6 CONCLUSION
The proposed algorithm, namely Robust Neighbourhood Search, is tested for altogether 1080
problem instances with a varying number of jobs, machines and setup times. The parameters
of the heuristics are optimized using design of experiment methodology. Further, experimen-
tations were carried out and makespan results were obtained. Three stages of analyses; graphi-
cal analysis, relative performance index analysis and statistical analysis are performed on the
results obtained. In the analyses, the proposed heuristic is compared with GA and are found
superior.
The main advantage of RNS is that the algorithm is very flexible and robust. It gives
improved result in successive iterations without getting trapped in local optima. The other
advantage is that the search techniques can be easily changed so as to adapt for real life situ-
ations. The local and global search parameters can also be changed. The statistical analysis
reveals that for some problem groups, the proposed algorithm is found to be not superior in
spite of the better makespan results obtained, which can be identified as shortcoming of the
proposed heuristic.
REFERENCES
Jatinder N.D & Gupta. 1986. The two machine sequence dependent flow-shop scheduling problem.
European Journal of Operational Research. 24(3): 439–446.
Laha D & Chakraborty UK .2009. A constructive heuristic for minimizing makespan in no-wait flow
shop scheduling. Int. j. Adv. Manuf. Technol. 41(1–2): 97–109.
Liu B. 2007. An effective PSO based algorithm for scheduling. IEEE 37(1): 45–52.
Malakooti B, Kim H & Shikh S .2011. Bat intelligence search with application to multi objective mul-
tiprocessor scheduling optimization. International journal for advanced manufacturing technology.
60(9): 1071–1086.
Marichelvan M. 2002. An improved hybrid Cuckoo Search (IHCS) meta heuristic algorithm for permu-
tation flow shop scheduling problems. International journal for bio inspired computation 4(4): 116–128.
Parthasarathy S, Rajendran CA .1997. A simulated annealing heuristic for scheduling to minimise
weighted tardiness in a flow shop with sequence dependent setup times of jobs—a case study, Produ
Plan Control 8(5): 475–483.
333

Rios-Mecardo R.Z. & Bard J.F. 1998. Heuristics for the flow line problem with setup cost. EJOR 110(1):
76–78.
Ruiz R., Maroto C. & Alcatraz J. 2005. Solving the flow shop scheduling problems with sequence
dependent setup times using advanced meta heuristic. EJOR, 165(1): 34–54.
Taillard E. 1989. Benchmark for basic scheduling problems. EJOR 64(2): 278–285.
Vanchipura, Rajesh & R. Sridharan, 2012. Development and analysis of constructive heuristics algo-
rithms for flow shop scheduling problems with sequence dependent setup times. International journal
for advanced manufacturing and technology. 67(5-8): 1337–1353.
Yagmahan B., Yensiey M. 2007. Ant colony optimization for multi-objective flow shop scheduling prob-
lem. Computers and Industrial Engineering. 54(3): 411–420.
334

A real-world analysis of the impact of knowledge management on

cost of quality in construction projects
Reshma Chandran & S. Ramesh Krishnan

Department of Mechanical Engineering, Rajiv Gandhi Institute of Technology, Kottayam, Kerala, India
ABSTRACT: Knowledge Management (KM) is vital in the case of the construction indus-
try because of its impact on integrating knowledge within and outside the industry. Knowl-
edge management implementation strategies play an important role in increasing project
performance. The need to reduce Cost of Quality (CoQ) is clear, but the effect of KM in low-
ering CoQ is uncertain. This paper reviews the factors contributing to CoQ and the effect of
KM on this. A literature survey and qualitative inquiries were adopted to address the research
aims. Organizations should understand that there is a native CoQ problem that needs to be
addressed. The main factors contributing to CoQ are design changes, errors and omissions,
and poor skills. Here, the logic is based on the desire to reduce CoQ and the need to tackle as
well as integrate knowledge across personal, project, organizational, and industry boundaries.
Knowledge management was found to have a positive impact on lowering the cost of quality.
Keywords: Real-world analysis, Knowledge Management, Cost of Quality
1 INTRODUCTION
Knowledge Management (KM) is the process of efficient handling of information and

resources within a commercial organization. Knowledge in an organization can be broadly
classified as explicit knowledge and tacit knowledge. The major region of focus is actually the
cost associated with the unwanted problem of redoing processes that have been inaccurately
implemented, which is often referred to as the cost of poor quality. This consists of costs
involving errors and omissions, poor skills, design changes, and the substantial costs often
associated with client disappointment.
The study focuses on KM aspects such as repetition of mistakes and an absence of lessons
learned, which may directly give rise to Cost of Quality (CoQ) issues.
2 LITERATURE REVIEW
2.1 Knowledge
Knowledge can be defined as the theoretical or practical understanding of a subject. A grouping
of information, background, and understanding can be captured, utilized and shared for business
purposes (Wibowo & Waluyo, 2015). Explicit knowledge is solitary that is able to be calculated,
taken into custody, examined, and can effortlessly be passed onto others in a codified layout. It
is that type of knowledge that can be uttered in statements of words and numbers. It can be sup-
plementary, transferred, dispersed and transformed in a methodical and prescribed way into facts
(Wibowo & Waluyo, 2015). Tacit knowledge, on the other hand, comes from one’s experience.
It can be considered as human knowledge that can be an intuition, finding, talent, experience, a
form of body language, or a value or belief. It is highly personal and context-specific, which is very
complicated to create, exchange in a few words, or distribute onwards to a community.
335

2.2 Knowledge management
Knowledge management was actually introduced more than two decades ago to help com-
panies generate, categorize, and utilize knowledge in an orderly manner. KM can be defined
as the classification, optimization and dynamic management of logical property to create
worth, increase efficiency, and gain benefit (Wibowo & Waluyo, 2015).
2.3 Issues associated with knowledge management in construction

Knowledge management is very useful to the construction industry as it is significant for construc-
tion organizations to bind together knowledge on the way to improve effectiveness and amplify
profitability. It is predominantly significant because of the unique nature of projects, such as the
difficult character of operations, professions and organizations, short-term team members, heavy
reliance on experience, rigid schedules, and restricted budgets (Nonaka & Konno, 1998).
3 RESEARCH METHODOLOGY
This paper reviews the factors contributing to the cost of quality and the effect of knowledge
management on the cost of quality. A literature survey and qualitative inquiries were adopted
to address the research aims.
The main factors contributing to the cost of quality are established using a literature survey.
Qualitative inquiries are used to identify the effect of knowledge management on the cost of qual-
ity. These are carried out using semi-structured interviews with open-ended questionnaires. For
the research, ten experts (see Table 1) were selected from construction companies across Thiru-
vananthapuram with knowledge management strategies. The questionnaire has three sections.
The first consists of general information, the second deals with the factors contributing to the
cost of quality, and the third section inquires into the effect of knowledge management on CoQ.
4 DATA COLLECTION AND ANALYSIS
Ten respondents were asked to rate the intensity of effect of KM processes on the compo-
nent elements of CoQ, that is, design changes, poor skills, and errors and omissions, based
on their own experience on construction projects. A four-point Likert scale was used to rate
the effect as follows: 1 = Strong Negative Impact; 2 = Negative Impact; 3 = Positive Impact;
4 = Strong Positive Impact. The mean values of the ratings were considered and the process
with most impact on CoQ in practice is determined.
Most of the respondents rated the KM processes as having a positive impact or strong
positive impact, that is, as 3 and 4. The data collected were analyzed and the leading five
Table 1. Profile of interviewees.
ID Years of experience Project experience* Organization experience*
A 36 1 1
B 27 1 1,2
C 24 1,2 1,2,3
D 12 1 1,2
E 14 1 1,2
F 12 1 1
G 10 1 1
H 13 1,2 1,2
I 12 1,2 1,2
J 14 1 1
*Key – Project experience: 1 – Building construction, 2 – Highway; Organization experience: 1 – Client

organization, 2 – Consultancy, 3 – Design.
336

Table 2. The impact of KM on errors and omissions.
N Min. Max. Mean
Knowledge champions 10 3 4 3.5

Knowledge capture 10 3 4 3.41
Knowledge creation 10 3 4 3.36
Knowledge transfer 10 3 4 3.34
Knowledge sharing 10 3 4 3.29
Table 3. The impact of KM on design changes.
N Min. Max Mean
Knowledge sharing (early) 10 3 4 3.55

Knowledge sharing (team) 10 3 4 3.37
Knowledge dissemination 10 3 4 3.21
Table 4. The impact of KM on poor skills.
N Min. Max. Mean
Knowledge identification 10 3 4 3.66

Knowledge transfer 10 3 4 3.51
Knowledge champions 10 3 4 3.39
Figure 1. Mean values of KM processes impacting errors and omissions.
KM processes that affect cost of quality were found by taking the mean of the values. This is
captured in Tables 2, 3 and 4, and also represented in the form of pie charts (see Figures 1, 2,
and 3). In the first case, that is, for the effect of KM processes on omissions and errors, the
responses obtained from the ten interviewees were 3, 3, 3, 3, 4, 4, 4, 4, 4 and 3. The average
value was 3.5. All other means were similarly calculated.
337

Figure 2. Mean values of KM processes impacting design changes.
Figure 3. Mean values of KM processes impacting poor skills.
5 FIVE HIGHEST-RANKING KM PROCESSES IMPACTING CoQ
The cost of quality mainly involves design changes, errors and omissions, and poor skills. The
major KM processes reported to affect errors and omissions are knowledge champions, knowl-
edge capture, knowledge creation, knowledge transfer, and knowledge sharing. Those affecting
design changes are knowledge sharing, knowledge creation, knowledge capture, and knowledge
dissemination. On the other hand, poor skills are mainly affected by knowledge identification,
knowledge transfer, knowledge capture, knowledge champions, and knowledge creation.
The main contributing factors to the cost of quality are errors and omissions, design changes,
and poor skills. These factors are summarized in Tables 5, 6 and 7.
According to the results, knowledge management has most impact in reducing errors
and omissions. The unifying process across all three CoQ contributing factors is knowledge
capture.
338

Table 5. Factors contributing to the cost of errors and omissions.
Factors contributing to the cost

of errors and omissions Description
Knowledge-related mistakes Insufficient idea about a task results in wrong solution

Rule-related mistakes Selecting an improper solution for a problem
Lack of lessons learned Due to insufficient learning from the past
Organizational culture Beliefs and norms which may lead to conflict or
opposition
Time constraints Errors made because workers are under time pressures
Budget constraints Limited budget leads to absence of good quality
resources
Poor communication This may include language barriers, improper transfer
of knowledge
Table 6. Factors contributing to the cost of design changes.
Factors contributing to the cost of design changes Description
Client change Client alteration to product requirements,

definition and project scope
Design change inevitable Design is actually an iterative practice, and is
expected to change during construction
Unforeseen site conditions Unpredicted ground conditions, weather, etc.
Poor client expertise Improper understanding of design scope
and cost
Errors and omissions in the site Non-reversible mistakes and omissions
resting on site results in alteration of
obtainable plan
Procurement strategy Those which do not involve design
innovation
Table 7. Factors contributing to the cost of poor skills.
Factors contributing to Description

the cost of poor skills
High personnel turnover There is a cycle of recruitment

Lack of dedication This is mainly with younger people, who are not prepared to
continue for a long time
Lack of training Industry lacks adequate training
Time constraints This may lead to inadequate transfer of knowledge because
workers are under time pressures
Organizational culture Insufficient knowledge distribution and knowledge conveyance
between project personnel
7 CONCLUSION
The main aims of this paper were to determine the factors contributing to the cost of qual-
ity and the effect of knowledge management on CoQ. It was found that the main factors
contributing to CoQ are errors and omissions, design changes, and poor skills. From the pie
charts, it is clear that KM processes such as knowledge sharing, knowledge creation, knowl-
edge capture, knowledge dissemination, knowledge transfer, and knowledge champions have
339

the highest impacts on cost of quality. Through qualitative inquiries, it was found that knowl-
edge management has a positive impact on reducing the cost of quality. Therefore, it is crucial
that every organization implement its own knowledge management strategies.
REFERENCES
Furcada, N. & Marcarulla, M. (2013). Knowledge management perceptions in construction and design
companies. Automation in Construction, 29, 83–91.
Gromoff, A., Kazantsev, N. & Bilinkis, J. (2016). An approach to knowledge management in construc-
tion service-oriented architecture. Procedia Computer Science, 96, 1179–1185.
Nonaka, I. & Konno, N. (1998). The concept of “Ba”: Building a foundation for knowledge creation.
California Management Review, 40(3), 40–54.
Wibowo, M.A. & Waluyo, R. (2015). Knowledge management maturity in construction companies.
Procedia Engineering, 125, 89–94.
340

Maintenance strategies for realizing Industry 4.0: An overview
A.A. Sambrekar, C.R. Vishnu & R. Sridharan

Department of Mechanical Engineering, NIT Calicut, Kerala, India
ABSTRACT: This paper presents an overview of different maintenance strategies widely

discussed in the industry by consolidating the results and inferences from a wide range of
research articles. This information will be useful to maintenance personnel during the selec-
tion and execution of maintenance activities to realize the concept of Industry 4.0. The paper
follows a combined meta-analytic and descriptive procedure for reviewing articles that have
provided path-breaking contributions in maintenance engineering and management. Details
of top-cited research articles related to maintenance are downloaded from the Web of Sci-
ence database to carry out the meta-analysis. A text analytic tool called BibExcel is utilized
for fetching and analyzing textual information. Subsequently, a close examination of signifi-
cant papers that describes different types of maintenance strategies is analyzed by following
the descriptive review procedure to report the advantages and limitations of each strategy.
Accordingly, it is found that predictive maintenance strategy is more aligned with the objec-
tives of Industry 4.0. Hence, an emphasis is placed on predictive maintenance techniques
and tools since this approach is gaining more attention as a result of advancements in soft
computation, cloud technology, data analytics, machine learning and artificial intelligence.
1 INTRODUCTION
Generally, maintenance involves the set of activities carried out in the industry to make sure
all the machineries and other physical assets are available for production. The main purpose
of industrial maintenance is to achieve minimum breakdown and to maintain efficiency of
the production facilities at the lowest possible cost. Maintenance activities and its execution
depend on the manufacturing system and layout of the plant. However, in any case, main-
tenance should not be considered as a cost-centric activity, but a profit-generating function
(Alsyouf 2007).
Maintenance helps in adding value to the organization through better utilization of pro-
duction facilities, enhancing product quality as well as reducing rework and scrap. ISO 55000:
2014 Asset Management System upholds, “Assets exist to provide value to the organization
and its stakeholders”. Unfortunately many companies still consider maintenance activities
as a “necessary evil”, due to the blurred perception about its role in attaining company’s
objectives and goals (Duffuaa et al. 2002). For those companies, the first step is to change
the corporate mindset such that the role of maintenance in achieving customer oriented per-
formance parameters such as quality, on-time delivery, etc., is significant.
Unexpected failures affect three key elements of competitiveness—quality, cost, and pro-
ductivity. In the modern world, all firms are striving hard to elevate these key features to
develop a strategic advantage against their competitors. Simply, waiting for the failure to
occur is not affordable in today’s business operations scene. Hence, companies have to adopt
different maintenance strategies suitable for their businesses.
The concept of Industry 4.0 originated from Germany, but its vision has caught the
attention of organizations across the globe (Zezulka et al. 2016). Industry 4.0 has spawned
a new wave of technology revolutionizing manufacturing environment through “smart
factory” in which machines cooperate with humans in real time via the cyber-physical
341

systems, thus creating a customer-oriented production field utilizing the technologies of
Internet-of-Things (IoT). The machine rather than working as an independent entity will
be able to collect data, analyze it and act accordingly. The use of advanced IoT sensors on
the machines has made it possible to detect any faults that go unseen by the human eye.
Instead of reacting to the failure, predictive maintenance will provide early warnings well
in advance so that machines or humans can take the necessary action to minimize the fre-
quency of failures (Tupa et al. 2017).
This paper presents an overview of different maintenance strategies widely discussed in
the literature. The paper has two main objectives. The first objective is to realize a systematic
review of literature in maintenance which is described in section 2. The second objective is to
explore how each of these maintenance strategies approaches toward maintenance decision
making which is described in section 3. Since predictive maintenance is widely being adopted
in today’s industry, section 4 explores the developments in this approach in maintenance.
2 META-ANALYSIS OF LITERATURE
A comprehensive amount of research papers have been published in the area of plant main-
tenance. It would be non-exhaustive process to review all the papers in the literature. The Web
of Science (WoS) database (after proper refinement) suggests 12,576 articles connected with
the keywords-Preventive Maintenance, Predictive Maintenance, Corrective Maintenance and
Reliability Centered Maintenance as on October, 2017. The collection includes 7,736 journal
articles, 4,744 conference proceedings and 609 reviews.
This huge volume of articles makes it difficult to provide a simple narration and critical
review of each research article. To overcome this common issue in conducting a literature
review, Glass et al. (1981) introduced a new approach known as Meta-analysis. Meta-analysis
is a methodology used to integrate research findings from a large body of articles using sta-
tistical analysis and sophisticated measurement techniques (Krishnaswamy et al. 2007). This
methodology is widely accepted in the research community to obtain firsthand information
from a large pool of articles from which the current direction of research can be projected
and significant articles in the domain can be shortlisted for further study.
To achieve the objective of this review paper, statistical techniques are employed to the
data retrieved from a pool of research papers extracted by data mining using BibExcel soft-
ware tool. Bibexcel is used to carry out bibliographic and statistical analysis by extracting
the data of textual nature such as title of the paper, author names, journal name, keywords,
etc. This free software tool also allows modifying and/or adjusting data that can be imported
from various databases including Scopus, WoS, Mendeley among others (Fahimnia et al.
2015).
Initially, a WoS outfile is created in plain text format that contains relevant information of
top 250 cited research papers in the maintenance domain, to be used as input for BibExcel.
The result of the analysis discloses the major contributors to the domain, major journals
publishing top quality articles, tools used in maintenance studies, etc.
2.1 Keyword frequency analysis

The occurrence of different research aspects can be related to the frequency of keywords
appeared in maintenance related articles as presented in the Figure 1.
It is evident from the figure that reliability is the most related aspect in industrial main-
tenance. Reliability has always been an important aspect in the assessment of industrial
equipment’s health. The choice of maintenance strategy highly influences the reliability of a
system. This is followed by optimization, preventive maintenance, fault diagnosis, condition
monitoring, etc. It is also inferred that genetic algorithm is the most used meta-heuristics to
solve optimization problems in maintenance. The title term frequency analysis conducted on
the same set of papers also substantiates similar results.
342

Figure 1. Keyword frequency analysis.
Figure 2. Significant researchers.
343

Table 1. Distribution of papers across journals.
Journal Number of papers
Reliability Engineering & System Safety 112

Expert Systems with Applications 28
European Journal of Operational Research 21
International Journal of Production Economics 13
IEEE Transactions 13
Computers & Operations Research 9
Journal of the Operational Research Society 8
Naval Research Logistics 8
Computers & Industrial Engineering 7
2.2 Analysis of significant researchers in the field of maintenance

The significant researchers who have published articles in the field of industrial maintenance
are represented in the Figure 2. It can be noted that Zio E., Berenguer C., Yang B.S., Coit
D.W. and Zuo M.J. have published substantial number of articles in the area of industrial
maintenance.
2.3 Journal frequency analysis

The majority of the top cited articles in industrial maintenance appeared in the journals such
as Reliability Engineering & System Safety, Expert Systems with Applications and European
Journal of Operational Research (Table 1). Therefore, these journals could be acknowledged
as leading ones in the areas of industrial maintenance.
3 MAINTENANCE STRATEGIES
In general, maintenance strategies can be generally classified into two categories: Reac-
tive and Proactive. Reactive maintenance focuses on repairing an asset once failure occurs.
Proactive maintenance focuses on avoiding repairs and asset failure through preventive and
predictive methods. These strategies meet the objectives laid out in the philosophy of Total
Productive Maintenance (TPM).
3.1 Corrective maintenance

Corrective maintenance (CM) also known as reactive maintenance refers to activities that
are done when equipment has already broken down, in order to restore the equipment to its
normal operating condition. The occurrence of sudden failure of the asset leads to high levels
of machine downtime. There is no initial cost associated with corrective maintenance and it
requires no planning. CM is suitable for non-critical components where the failure neither
compromises employee safety nor the production objectives.
3.2 Scheduled maintenance

Scheduled maintenance (or preventive maintenance) refers to the set of activities that are
performed at regular intervals on a system to reduce the likelihood of it failing. Usually, pre-
ventative maintenance is carried out with minimal disruption in production activities such
that unexpected breakdowns will not occur. The maintenance is scheduled either time based
or usage based. An air-conditioner which is serviced every year, before summer is an example
of time based preventative maintenance whereas, the maintenance of motor vehicles are usu-
ally carried out on a usage based schedule (e.g. every 10,000 km).
344

3.3 Predictive maintenance
Machines usually undergo degradation before failure occurs. It is possible to monitor the trend
of degradation so that any faults can be corrected before they cause any failure and machine
breakdown. Predictive maintenance (PdM) is one such strategy that helps us to predict failures
before they actually occur. It is a more efficient strategy since maintenance is only performed
on machines when it is required. This strategy requires condition monitoring of the asset that
detects signs of decreasing performance or upcoming failure by utilizing sensor technologies.
The asset is monitored for various parameters such as vibration, temperature, lubricating
oil, contaminants, and noise levels. The general classification of maintenance strategies is
provided in Figure 3.
Maintenance decision making under the PdM can be analyzed in two perspectives namely:
diagnosis and prognosis. Diagnosis is the process of identifying the root-cause of a fault
(Jeong & Villalobos 2007), while prognosis is the process of predicting failure that may occur
in future (Lewis & Edwards 1997). Diagnostics comprises (a) fault detection, (b) fault isola-
tion and (c) fault identification, while prognostics include (a) remaining useful life (RUL)
prediction and (b) confidence interval estimation (Efthymiou et al. 2012). Even if the asset is
running in a degraded state, it cannot be concluded that the asset has failed. It can probably
still be utilized for certain duration before failure occurs. To deal with these circumstances,
prognosis is required. The key purpose of prognosis is to provide early warning to facilitate
better maintenance planning. Therefore, appropriate maintenance activities for the equip-
ment can be planned to prevent failures.
3.4 Reliability-centered maintenance

Reliability-Centered Maintenance (RCM) is a systematic approach for determining the most
effective approach for maintenance. Effectiveness is determined by considering either reliabil-
ity (or probability of failure) and overall cost. RCM is one of the best known and most used
techniques to preserve the operational efficiency in critical sectors like power plants, artillery
system, aviation industry, railway networks, oil and gas industry and ship maintenance (Car-
retero et al. 2003).
Generally, RCM is conceived as the optimum mix of all the traditional maintenance strategies
in order to utilize the advantages of those maintenance strategies. While preventive maintenance
is generally considered to be worthwhile, there are some disadvantages such as huge cost and
need of specialist labor. Hence, preventive maintenance need not be the cost effective strategy for
every machinery/component especially for the non-critical assets that every industry possesses.
Thus, for proper maintenance of the plant, it is better to adopt an integrated method
of breakdown and preventive maintenance strategies to make use of the respective strength
Figure 3. Classification of maintenance strategies.
345

alone for which RCM was introduced. Accordingly, RCM selects the most appropriate and
tailor made maintenance strategy for all the equipment in the plant based on its criticality
score and reliability parameters (Vishnu & Regikumar 2016).
3.5 Risk based maintenance

Risk Based Maintenance (RBM) prioritizes maintenance resources toward assets that carry
the most risk if they were to fail. Risk-based maintenance framework comprises two major
phases: (Selvik & Aven 2011, Arunraj & Maiti 2010)
i. Risk assessment
ii. Maintenance planning based on risk
Risk is adopted as an index for clarifying priority in risk maintenance technologies. Research-
ers have also proposed risk based models for optimizing maintenance activities mainly using
the tools of simulation (Nielsen & Sørensen 2011, Sørensen 2009). As per Khan & Haddara
(2003), risk is be calculated as the product of failure probability and consequences of failure.
Generally, it is conceived that 20% of the components in the system will comprise 80%
of the total risk of the system. In that case, it would be irrational to inspect all devices with
the same level of priority. It is thus important to identify the 20% of devices and raise their
priority level in the inspection program. This is the basic concept of risk based maintenance
strategy. This concept is called the 80–20 rule (Pareto principle).
3.6 Design-out maintenance

Design-out maintenance focuses on the vulnerabilities in the machines that lead to fre-
quent failures and aims to minimize the regular maintenance activities by redesigning those
machines and facilities. This will raise machine performance from the maintenance point of
view, especially to those machines having longer repair time or huge replacement cost.
4 DEVELOPMENTS IN PREDICTIVE MAINTENANCE
Predictive Maintenance is the current excitement in the field of maintenance. This strategy is
based on maintenance predictive models working on the principle of machine learning. When
building predictive models, we use historical data to train the model which can then recognize
hidden patterns and further identify these patterns in the future data. These models are trained
with examples described by their features and the target of prediction. The trained model is
expected to make predictions on the target by analyzing the features of the new examples. It is
crucial that the model captures the relationship between features and the target of prediction.
In order to train an effective machine learning model, we need training data which includes
features that actually have predictive power towards the target of prediction.
Therefore, quantitative data is of foremost importance for training the model. Since fail-
ures are very rare to occur and the availability of such data is difficult in real time, there
are a few data repositories like Prognostics and Health Management (PHM) that facilitate
the availability of run-to-failure data to encourage development of prognostic algorithms.
Saxena et al. (2008) have described how damage propagation can be modeled in various
modules of aircraft gas turbine engines for developing and testing prognostics algorithms.
This paper also presents an evaluation policy for performance benchmarking of different
prognostics algorithms.
In Prognostics, the objective is to predict the Remaining Useful Life (RUL) before a failure
occurs, given the current machine condition and past operation profile. Since the dataset
was made available by PHM in 2008, researchers have built different prognostics methods.
Ramasso & Saxena (2014) reviewed different approaches to PHM dataset and analyzed to
understand why some approaches worked better than others. This paper presents three top
winning approaches described in the order of their rank as follows:
346

• Wang et al. (2008) present a similarity-based approach for estimating the RUL in prognostics,
however, it was only suitable when abundant amount of run-to-failure data was available.
• Heimes (2008) utilizes advanced recurrent neural network architecture to estimate the
RUL of the system,
• Peel (2008) utilizes ensemble regression modelling for determination of RUL.
Different other approaches proposed by Fagogenis et al. (2015) and Bucknam (2017)
have also been utilized for predicting the RUL of the aircraft engine for the PHM data-
set after 2008. The results of these algorithms are significantly closer to the three winning
approaches.
The value of predictive maintenance is already well recognized by the biggest players in the
industry. Rolls Royce adopted predictive maintenance strategy under its Total Care Program
which resulted in transforming them from a loss making aircraft manufacturer to a leading
aircraft engine manufacturer. The change of maintenance strategy highly impacted the busi-
ness decisions for Rolls-Royce and the company has now moved to leasing its engines in the
monthly subscription model. They have added complex system monitoring sensors in its
engines that sends real-time engine performance metrics when the plane is mid-air to one of
the company’s R&D centers. The data is then analyzed immediately and any concerns will be
conveyed to the aircraft to ensure its safety.
5 SCOPE FOR FUTURE WORK
The previous discussion reveals that although the application of PdM is more beneficial com-
pared to other strategies from a practical point of view, further research on PdM is necessary
in order to realize the concept of Industry 4.0. The application of PdM is more complex and
expensive because PdM heavily relies on the implementation of complex sensor systems to
monitor the health conditions of the equipment in real time. As discussed, research based
on data repository provided by Prognostic and Health Management (PHM) seems the right
direction for future research on condition based maintenance of sensitive and complex equip-
ment/systems. As observed from the literature, around 70 publications have been found that
have utilized the PHM dataset for the development of prognostic algorithms. However, the
PHM dataset exhibits an exponential degradation pattern. No comparison has been made to
check whether the same prognostic algorithms can perform effectively on different datasets.
6 CONCLUSION
This paper presented a systematic review of different maintenance strategies. Although many
papers have been published in this area, only a few papers present the advantages and chal-
lenges associated with implementing each maintenance strategy. This paper identifies cer-
tain ‘core’ articles which may prove beneficial for the people seeking to research the area.
We identified some of the recent path-breaking research papers that contribute towards the
advancements in maintaining production facilities and other complex machinery to achieve
efficiency and effectiveness.
Furthermore, it can be seen that there have been tremendous excitements all over the world
to implement predictive maintenance strategy for monitoring and executing maintenance
activities. This motivation is associated with the advancements in the field of systems engi-
neering especially in machine learning, data analytics, and instrumentation technologies.
Hence, PdM can be seen as the most promising maintenance strategy for realizing the objec-
tives of Industry 4.0. However, a single maintenance strategy cannot be the most economical
strategy for all the equipment in the plant. Especially for non-critical items, it is always better
to follow breakdown maintenance. Therefore, it can be concluded that an optimum mix of
the above strategies such as RCM, with more emphasis on predictive maintenance, is the
most suited maintenance strategy in the emerging industrial scenario.
347

REFERENCES
Alsyouf, I. 2007. The role of maintenance in improving companies’ productivity and profitability. Int.
J. Prod. Eco. 105(1): 70–78.
Arunraj, N.S. & Maiti, J. 2010. Risk-based maintenance policy selection using AHP and goal program-
ming. Safety Science, 48(2): 238–247.
Bucknam, J.S. 2017. Data analysis and processing techniques for remaining useful life estimations.
Carretero, J., Pérez, J.M., Garc a-Carballeira, F., Calderón, A., Fernández, J., Garc a, J.D., Lozano, A.,
Cardona, L., Cotaina, N. & Prete, P., 2003. Applying RCM in large scale systems: a case study with
railway networks. Relia. Engng. & System Safety 82(3): 257–273.
Duffuaa, S.O., Al-Ghamdi, A.H. & Al-Amer, A. 2002. Quality function deployment in maintenance
work planning process. In Proc. of the 6th Saudi engineering conference KFUPM, Dhahran, Kingdom
of Saudi Arabia.
Efthymiou, K., Papakostas, N., Mourtzis, D. & Chryssolouris, G. 2012. On a predictive maintenance
platform for production systems. Procedia CIRP, 3: 221–226.
Fagogenis, G., Flynn, D. & Lane, D., 2014. Novel RUL prediction of assets based on the integration of
auto-regressive models and an RUS Boost classifier. Prognostics and Health Management (PHM),
2014 IEEE Conference: 1–6.
Fahimnia, B., Sarkis, J. & Davarzani, H. 2015. Green supply chain management: A review and biblio-
metric analysis. Int. J. Prod. Eco.162: 101–114.
Glass, G.V., McGaw, B., Smith, M.L. 1981. Meta Analysis in Social research. Beverly Hills, CA: Sage
Publications, New York, NY.
Heimes, F.O., 2008. Recurrent neural networks for remaining useful life estimation. In Prognostics and
Health Management (PHM), 2008 IEEE Conference: 1–6.
Jeong, I.J., Leon, V.J. & Villalobos, J.R. 2007. Integrated decision-support system for diagnosis, mainte-
nance planning, and scheduling of manufacturing systems. Int. J. Prod. Research, 45(2): 267–285.
Khan, F.I. & Haddara, M.M. 2003. Risk-based maintenance (RBM): a quantitative approach for
maintenance/inspection scheduling and planning. J. of loss prevention in the process industries, 16(6):
561–573.
Kim, N.-H., Joo-Jo, C. & An, D. 2017. Prognostics and Health Management of Electronics.
Krishnaswamy, K.N., Sivakumar, A.I., Mathirajan, M. 2007. Management Research Methodology,
Dorling Kindersley: Pearson India Publications, India.
Kuhn, M. & Johnson, K. 2013. Applied predictive modeling (Vol. 810). New York: Springer.
Lewis, S.A. & Edwards, T.G. 1997. Smart sensors and system health management tools for avionics and
mechanical systems. 16th DASC. AIAA/IEEE Digital Avionics Systems Conference. Reflections to the
Future. Proceedings 2: 8.5–1−8.5–7. doi: 10.1109/DASC.1997.637283.
Nielsen, J.J. & Sørensen, J.D. 2011. On risk-based operation and maintenance of offshore wind turbine
components. Relia. Engng. & System Safety, 96(1): 218–229.
Peel, L. 2008. Data driven prognostics using a Kalman filter ensemble of neural network models. 2008
Int. Conference on Prognostics and Management 1–6, doi: 10.1109/PHM.2008.4711423.
Ramasso, E. & Saxena, A. 2014. Performance Benchmarking and Analysis of Prognostic Methods for
CMAPSS Datasets. Int. J. Prognostics and Health Management 5(2):1–15.
Saxena, A., Goebel, K., Simon, D. & Eklund, N. 2008. Damage Propagation Modeling for Aircraft
Engine Prognostics. Response.
Selvik, J.T. & Aven, T. 2011. A framework for reliability and risk centered maintenance. Relia. Engng.
and System Safety 96(2):324–331.
Si, X.S., Zhang, Z.X. & Hu, C.H., 2017. Data-Driven Remaining Useful Life Prognosis Techniques:
Stochastic Models, Methods and Applications. Springer.
Sørensen, J.D. 2009. Framework for risk-based planning of operation and maintenance for offshore
wind turbines. Wind energy, 12(5): 493–506.
Tupa, J., Simota, J. & Steiner, F. 2017. Aspects of risk management implementation for Industry 4.0,
Procedia Manufacturing 11, 1223–1230.
Vishnu, C.R. & Regikumar, V. 2016. Reliability Based Maintenance Strategy Selection in Process Plants:
A Case Study. Procedia Technology 25:1080–1087.
Wang, T., Yu, J., Siegel, D. & Lee, J., 2008. A similarity-based prognostics approach for remaining useful
life estimation of engineered systems. In Prognostics and Health Management (PHM) 2008. IEEE
Conference:1–6.
Zezulka, F., Marcon, P. Vesely, I. & Sajdl, O. 2016. Industry 4.0 – An Introduction in the phenomenon.
IFAC-Papers Online, 49(25), 8–12.
348

Manufacturing technology and material science

Characterization of nanoliposomes and their modification for

drug delivery
K.S. Athira
Department of Materials Engineering, Indian Institute of Science, Bangalore, Karnataka, India
K.W. Ng
School of Materials Science and Engineering, Nanyang Technological University, Singapore
ABSTRACT: Atherosclerosis causes heart disease and stroke and is a main cause of death
in many countries. Nanoliposomes are a potential candidate for use in drug targeting for its
treatment. The size of the nanoliposomes affects their cellular uptake. An optimal size range
is necessary for effective site-specific drug delivery. Such optimum-sized nanoliposomes were
developed. Their sizes were characterized in this study using dynamic light scattering and
nanoparticle tracking analysis. The sizes of the nanoliposomes were found to be in the ranges
of 79–128 nm via the former method and 86–99 nm via the latter. Liposomes grafted with
polyethylene glycol show an improved stability and increased circulation time, while those
grafted with fluocinolone acetonide help to reduce inflammation. It was observed that nor-
mal nanoliposomes had a larger size than these grafted liposomes, which each had a larger
size than those grafted with both polyethylene glycol and fluocinolone acetonide.
Keywords: nanoliposomes, polyethylene glycol, fluocinolone acetonide, drug delivery
1 INTRODUCTION
Atherosclerosis is a chronic inflammatory disease in which endothelial stress or injury causes

low-density lipoproteins (LDLs) to infiltrate into the tunica intima of the artery. These LDLs
may undergo oxidization, and in oxidized form they are toxic to the endothelial cells. The
inflamed cells will then recruit monocytes (Wick et al., 2004). They migrate into the tunica
intima where they differentiate into macrophages. The macrophages take up oxidized LDLs
that permeate into and transform to foam cells (Gerrity, 1981). They aggregate to form an
atheromatous core, which later becomes necrotic (Ross, 1993).
Liposomes are small artificial vesicles of spherical shape that can be created from cholesterol
and phospholipids (Akbarzadeh et al., 2013). They have one or more concentric spherical lipid
bilayers with aqueous phases inside and between lipid bilayers. Nanoliposomes are a new tech-
nology for the encapsulation and delivery of bioactive agents. They have potential applications
in nanotherapy (diagnosis, cancer and atherosclerosis therapy, gene and drug delivery) because
of their biocompatibility, biodegradability and nanosizing. They can upgrade the functioning
of bioactive agents by enhancing their dissolvability and bioavailability, in vitro and in vivo sta-
bility, and, in addition, blocking their undesirable interactions with other molecules. Another
preferred application of nanoliposomes is cell-specific targeting, which is essential to achieving
the concentrations of drugs required for ideal therapeutic value at the target site while limiting
antagonistic impacts on healthy cells and tissues. Water-soluble compounds can be trapped in
the aqueous phase, and lipophilic agents can be trapped between liposomal bilayers (Torchilin,
2005). Liposomes naturally target the Mononuclear Phagocytic System (MPS), that is, mono-
cytes, macrophages and dendritic cells, and thus they can be used to deliver drugs to the MPS.
The size, charge and lipid composition affect the efficiency of liposomes targeting MPS cells
351

(Kelly et al., 2010). Characterization of the liposomes in terms of their size is very important
because the size of the drug will affect cellular uptake. Liposomes can be modified by grafting
with Polyethylene Glycol (PEG) or Fluocinolone Acetonide (FA).
Because PEG is an exceptionally hydrophilic polymer and has low toxicity, PEG and its
subsidiaries have been generally used to enhance the stability and pharmacokinetics of drug
carriers and parent drugs (Harris et al., 2001). In liposomal drug delivery, PEG has been
broadly utilized for liposome surface alteration (PEGylation), and this procedure has been
utilized for making liposomal drug delivery systems, which are known as PEGylated lipo-
somes (Santos et al., 2007). PEG also increases the circulation time (Bergstrand, 2003), which
is necessary for an efficient site-specific drug delivery.
Fluocinolone acetonide is an anti-inflammatory drug. The results of Vafaei et al. (2015)
reveal that the liposomes have promising potential as an effective delivery system for incor-
poration of fluocinolone acetonide following the formation of inclusion complex to release
the drug in a sustained manner for the treatment of ocular inflammatory disease. It can thus
be used to control inflammation in atherosclerosis.
Fluocinolone acetonide-loaded PEGylated liposomes have a strong pharmacological effect
and low toxicity (Vafaei et al., 2015). Hence, liposomes modified by grafting with PEG or FA
or both also need to be characterized as they provide a better drug delivery system.
It has been shown that liposomes in the size range of 50 to 800 nm show a trend in which
the small liposomes are optimal for internalization by the MPS cells, while the large ones
induce toxicity and cytokine activation, which leads to inflammation (Epstein-Barash et al.,
2010; Takano et al., 2003). Increasing liposome size resulted in increased inhibitory effect of
alendronate liposomes, liposomal clodronate and with h-monocytes and J774 macrophage
cell lines (Epstein-Barash et al., 2010). Reactive oxygen species generation was decreased by
PEG coating as the association with macrophage-like RAW264.7 cells and the induction of
apoptosis were reduced (Takano et al., 2003).
In this article, we have used two different methods to analyze the size characterization of
nanoliposome materials, together with their PEG coating material, the loaded drug and, finally,
the combined material, as effective surface modifications of nanoliposomes for site-specific drug
delivery in the treatment of atherosclerosis: the results obtained are compared and discussed.
Four types of nanoliposome were procured from the collaborative research program between
the School of Materials Science and Engineering, Nanyang Technological University,
Singapore, and other international universities. The four types of liposomes were:
a. Normal liposomes (blank);
b. Liposomes grafted with PEG;
c. FA-loaded liposomes;
d. Liposomes grafted with PEG-FA.
2.1 Dynamic light scattering

The size of particles suspended in a solution can be determined by measuring the changes in
the intensity of light scattered from the solution; this method is called Dynamic Light Scat-
tering (DLS).
The principle of dynamic light scattering is that fine particles that are in Brownian motion
(constant and random thermal motion) diffuse with a speed that is related to their size.
Smaller particles diffuse faster than larger ones. To measure this diffusion speed, the pattern
produced by illuminating the particles with a laser is observed.
The four samples were diluted to 10X, 100X, 1000X and 100,000X. The size of each sam-
ple was measured by the DLS method using a Zetasizer Nano ZS90 (Malvern Instruments
Ltd, UK) at 25°C.
352

2.2 Nanoparticle tracking analysis
In Nanoparticle Tracking Analysis (NTA), the size of the suspended particles in a solution
is found by measuring the rate of Brownian motion, which is correlated to the size of these
particles.
Similarly to DLS, the particles of the sample can be visualized by the light they scatter
when illuminated by a laser. The light scattered by the particles is captured using a camera
and the motion of each particle is tracked frame to frame by software. This rate of particle
movement is related to a sphere-equivalent hydrodynamic radius as calculated through the
Stokes–Einstein equation.
The four samples were diluted to 10,000X & 100,000X and the size was measured using a
Nanosight NS300 (Malvern Instruments Ltd, UK) at 25°C.
Representative size data for the four types of nanoliposomes from the DLS measurement is
shown in Figure 1. It can be seen from the figure that the size of the normal (blank) liposome is
approximately 100 nm. Similarly, data were obtained for all the different liposomes in each and
every dilution. The variation in the sizes of the different types of liposomes at varying dilutions
found by the DLS method is shown in Figure 2. The size of the nanoliposomes can be seen to be
in the range of 79–140 nm. The blank liposomes have a size range of 94–128 nm, with an aver-
age of 111 nm. The PEG-coated liposomes have a size range of 96–124 nm, with an average of
110 nm. The FA-loaded liposomes have a size range of 91–97 nm, with an average of 94 nm. The
PEG-FA-incorporated liposomes have a size range of 79–140 nm, with an average of 110 nm.
As the dilution of 10,000X is at the limit of the sensitivity of the DLS equipment, there is a large
amount of error in the data obtained at this dilution. So, excluding this, the average sizes of the
liposomes are plotted in Figure 3. Thus, the size of different types of nanoliposomes ranges from
79 to 128 nm, with an average of 104 nm. The trend of sizes from DLS measurement can be
described as: blank liposomes > PEG liposomes > FA liposomes > PEG-FA liposomes.
Representative size data from the NTA measurement is shown in Figure 4, in which the
concentration of particles of a particular size is plotted. In the given data, most of the parti-
cles have a size of 100 nm. Further, the small peaks of higher sizes can be ignored because of
the very small number of particles in that size range.
The average of the size values thus obtained from the dilutions 10,000X and 100,000X
is shown in Figure 5. The size of the nanoliposomes were found to be in the range of
86–99 nm. The trend of sizes obtained from NTA measurement is: blank liposomes > PEG
liposomes ≈ FA liposomes > PEG-FA liposomes.
Figure 1. Representative data of dynamic light scattering: Intensity of scattered light with size of
nanoliposomes (blank).
353

Figure 2. Size of nanoliposomes by dynamic light scattering.
Figure 3. Average sizes of the different types of nanoliposomes obtained by the dynamic light scat-
tering method.
Figure 4. Representative data for nanoparticle tracking analysis: Concentration of nanoliposomes

versus size.
354

Figure 5. Size of nanoliposomes by nanoparticle tracking analysis: Average of 10,000X and
100,000X.
Because NTA is a more accurate method, the overall trend obtained by combining the
results from DLS and NTA measurements can be summarized as: blank liposomes > PEG
liposomes ≈ FA liposomes > PEG-FA liposomes.
Larger liposomes exhibited increased internalization by h-monocytes (Epstein-Barash
et al., 2010), but with a higher degree of apoptosis induction (Takano et al., 2003). A sig-
nificant reduction in activity was also found after treatments with very small liposomes
(55 ± 15 nm) (Epstein-Barash et al., 2010). Hence, a size range between these two extremes,
such as 75–150 nm, would be ideal for a successful site-specific drug delivery.
It has been observed by Nicholas et al. (2000) that the addition of PEG or FA brought the
weight ratio of liposomes down due to permeabilities, reaction temperatures, and phase tran-
sition between the mushroom and brush regimes. This might be the reason for the reduction
in size observed in the liposomes grafted with PEG, FA, and both.
4 CONCLUSIONS
The size characterization study of the different types of nanoliposomes procured from the
collaborative research program between Nanyang Technological University, Singapore, and
other international universities was done using DLS and NTA methods.
The size of the nanoliposomes alone, as well as in forms grafted with different materials,
were found to be in the range of 79–128 nm via the DLS method and 86–99 nm via the NTA
method adopted in the present study.
Because all the samples were monodispersed, both techniques are appropriate and the
results are comparable. The standard deviation in the DLS method was 30.5 and that with
NTA was only 6.5. Hence, the NTA method is adjudged as the better method as it produces
less deviation in the data.
The trend of size variations in the materials used in the current study can be described as:
blank liposomes > PEG liposomes ≈ FA liposomes > PEG-FA liposomes. Thus, the nanoli-
posomes grafted with the two different materials necessary for drug delivery were found to be
better than blank nanoliposomes with respect to size. The PEG-FA grafted nanoliposomes
characterized with a size of 85.5 nm procured from this laboratory are found to be superior
in site-specific drug delivery for atherosclerosis with respect to size.
355

REFERENCES
Akbarzadeh, A., Rezaei-Sadabady, R., Davaran, S., Joo, S.W., Zarghami, N., Hanifehpour, Y. &
Nejati-Koshki, K. (2013). Liposome: Classification, preparation, and applications. Nanoscale
Research Letters, 8(1), 102.
Bergstrand, N. (2003). Liposomes for drug delivery: From physico-chemical studies to applications
(Doctoral dissertation, Acta Universitatis Upsaliensis, Sweden).
Epstein-Barash, H., Gutman, D., Markovsky, E., Mishan-Eisenberg, G., Koroukhov, N., Szebeni, J.
& Golomb, G. (2010). Physicochemical parameters affecting liposomal bisphosphonates bioactivity
for restenosis therapy: Internalization, cell inhibition, activation of cytokines and complement, and
mechanism of cell death. Journal of Controlled Release, 146(2), 182–195.
Gerrity, R.G. (1981). The role of the monocyte in atherogenesis: I. Transition of blood-borne mono-
cytes into foam cells in fatty lesions. American Journal of Pathology, 103(2), 181–190.
Harris, J.M., Martin, N.E. & Modi, M. (2001). Pegylation. Clinical Pharmacokinetics, 40(7), 539–551.
Kelly, C., Jefferies, C. & Cryan, S.A. (2010). Targeted liposomal drug delivery to monocytes and macro-
phages. Journal of Drug Delivery, 2011, 727241.
Nicholas, A.R., Scott, M.J., Kennedy, N.I. & Jones, M.N. (2000). Effect of grafted polyethylene glycol
(PEG) on the size, encapsulation efficiency and permeability of vesicles. Biochimica et Biophysica
Acta (BBA) - Biomembranes, 1463(1), 167–178.
Ross, R. (1993). The pathogenesis of atherosclerosis: A perspective for the 1990s. Nature, 362(6423),
801–809.
Santos, N.D., Allen, C., Doppen, A.M., Anantha, M., Cox, K.A., Gallagher, R.C., ... Webb, M.S.
(2007). Influence of poly (ethylene glycol) grafting density and polymer length on liposomes: Relat-
ing plasma circulation lifetimes to protein binding. Biochimica et Biophysica Acta (BBA) - Biomem-
branes, 1768(6), 1367–1377.
Takano, S., Aramaki, Y. & Tsuchiya, S. (2003). Physicochemical properties of liposomes affecting apop-
tosis induced by cationic liposomes in macrophages. Pharmaceutical Research, 20(7), 962–968.
Torchilin, V.P. (2005). Recent advances with liposomes as pharmaceutical carriers. Nature Reviews Drug
Discovery, 4(2), 145–160.
Vafaei, S.Y., Dinarvand, R., Esmaeili, M., Mahjub, R. & Toliyat, T. (2015). Controlled-release drug
delivery system based on fluocinolone acetonide–cyclodextrin inclusion complex incorporated in
multivesicular liposomes. Pharmaceutical Development and Technology, 20(7), 775–781.
Wick, G., Knoflach, M. & Xu, Q. (2004). Autoimmune and inflammatory mechanisms in atherosclero-
sis. Annual Review of Immunology, 22, 361–403.
356

Investigation of compressive strength of impact damaged hybrid

composite laminate
A. Madhan Kumar, V. Kathiresan, K. Ajithkumar, A. Hukkim Raja &

D. Balamanikandan
Coimbatore, India
ABSTRACT: The objective of this research work is to investigate the compressive strength
of Glass-Carbon/epoxy hybrid laminate subjected to impact damage. Glass-Carbon/epoxy
hybrid laminate was fabricated using vacuum assisted compression molding and a novel
arrangement of quasi-isotropic sequence was followed. Coupon specimens were prepared
according to ASTM standard for low velocity drop impact and Compression After Impact
(CAI) to assess its compressive strength after the impact. Results showed that the stacking
sequence has minimized the impact damage area. The failure of the laminate after CAI was
majorly due to the buckling of the sub laminate. The Hybridized effect played a vital role in
the performance of the laminate.
1 INTRODUCTION
Low velocity impact induced damage on composite structures is a dangerous phenomenon,

since it barely leaves any indication on the surface. Propagation of this internal damage can
lead to catastrophic failure of the complete structure. Moreover, the compressive strength
of the structure is reduced significantly. The internal damage, induced by the impact causes
matrix cracks, fiber breakage and delamination in the structure. Brittle nature of the matrix
is a crucial factor for delamination. Compression after Impact (CAI) strength is affected by
matrix cracks, matrix and fiber micro cracking, weakening of interlaminar strength as shown
by Hao Yan et al. (2010). Slattery, P.G. et al. (2016), Hao-Ming Hasiao. (2012), Yuichiro
et al. (2008), V. Kostopoulos et al. (2010), Xinguang Xu et al. (2014), Daniele Ghelli &
GiangiacomoMinak. (2011), Samuel Rivallant et al. (2013). Hakim Abdulhamid et al. (2016),
Jun-Jiang Xiong et al. (2008) carried out investigations with Carbon fiber [2–10] and Hao
Yan et al. (2010), Jefferson Andrew J. et al. (2015). Mehmet Aktas et al. (2012), Mandar D.
Kulkarni et al. (2011), Mehmet Aktas et al. (2009) investigated with Glass fiber. As Hybridi-
zation has evolved in recent years, it is important to study the hybridized effect of Glass and
Carbon.
Many researches were done on CAI tests. CAI simulation works were carried out by
Hakim Abdulhamid et al. (2016), Panettieri E. et al. (2016), Gonzalez E.V. et al. (2012), Tho-
mas E. Lacy & Youngkeun Hwang. (2003), Wei Tan et al. (2015). Experimental CAI works
were done by Slattery, P.G. et al. (2016), Hao-Ming Hasiao. (2012), Yuichiro et al. (2008),
V. Kostopoulos et al. (2010), Xinguang Xu et al. (2014), Tan K.T. et al. (2012), Mannov E.
(2013), Hakim Abdulhamid et al. (2016) and Sanchez Saez S. et al. (2005) on FRPs. CAI
on sandwich panels were carried out by Thomas E. Lacy & Youngkeun Hwang. (2003), Bin
Yang et al. (2015), Guoqi Zhang et al. (2013), Davies G.A.O. et al. (2004), Bruno Castanie
et al. (2008), Gilioli A. et al. (2014), Vaidya U.K. et al. (2000), Michael W. Czabaj et al. (2010)
and Andrey Shipsha & Dan Zenkert. (2005). Modifications were brought in through repair
by Slattery, P.G. et al. (2016) and Jefferson Andrew J. et al. (2015), stitch by Alaattin Aktas
et al. (2014), Aymerich F. & P. Priolo. (2008), Tan K.T. et al. (2015) and Tan K.T. et al. (2012),
357

pinning by Zhang X. et al. (2006) and Vaidya U.K. et al. (2000), fillers by V. Kostopoulos
et al. (2010), Xinguang Xu et al. (2014) and Mannov E. (2013), Tapered laminates byHakim
Abdulhamid et al. (2016), repeated impact by Cao D F et al. (2015) and Hakim Abdulhamid
et al. (2016) and through hygrothermal environment by Yuichiro et al. (2008), Mehmet Aktas
et al. (2009), Berketis K. & D. Tzetzis. (2010) and Michael Dale et al. (2012). However CAI
on Hybrid FRPs is meager. This paper is focused on the hybrid laminate.
2 MATERIALS AND METHOD
2.1 Fabrication
Seven layers of Plane weave Carbon (warp)-Glass (weft) hybrid (C-G) and six layers of Plane
weave Glass (G) (warp and weft) fiberswere selected for the laminate fabrication. Fibers were
cut to a size of 550 mm*400 mm. A total of 13 layers were taken for the laminate. A novel
quasi isotropic sequence was selected as shown in Figure 1 (a). Epoxy resin LY556 and hard-
ener HY952 were used in the ratio 10:1. Vacuum assisted compression molding was used to
fabricate the laminate. A pressure of 600 Pa was applied over the laminate for 4 hour and
cured at a temperature of 120°C for 12 hours and was left for ambient cure for 2 hours.
Figure 1 (b) shows the fabricated laminate of thickness 4.2 mm. Coupon specimens were pre-
pared according to ASTM standards to determine the basic properties and CAI properties.
2.2 Experiments
2.2.1 Basic properties
Uniaxial tensile test was performed to determine the basic properties. Samples were pre-
pared according to ASTM D3039. Table 1, shows the basic properties of Carbon-Glass/
epoxy hybrid laminate.
2.2.2 Low velocity impact and CAI test

Samples were prepared according to ASTM D7137. Coupon specimen were prepared for
the impact and CAI test. The impact test setup consists of a drop tower with hemispherical
impactor of mass 3.5 kg and 15.3 mm diameter. The impactor, when released from a height,
Figure 1. (a) Stacking sequence. (b) Fabricated laminate.
Table 1. Basic Properties of Glass-Carbon/epoxy hybrid laminate.
E1, GPa E2, GPa G12, GPa ν12
Glass-Carbon/epoxy hybrid laminate 14.32 11.55 4.493 0.5
358

impacts the specimen which was clamped below, with the corresponding energy. Energy level
can be varied by varying the height of the impactor on the drop tower. Three samples for
each of 25J, 35J and 45J were chosen. Compression tests were carried out for the impacted
samples in CAI fixture. The fixture consists of anti-buckling plates to prevent the buckling
of the sample during CAI test. The CAI was linked with the data acquisition system, from
which results were obtained.
Figures 2–4 shows the samples of impact energies of 25J, 35J and 45J respectively. For 25J
and 35J, visible damage area was observed. However,the sample of 45J energy,perforation
was observed. Matrix cracking has taken place at 25J. A reduction in the area of impact
damage was observed as compared to the work of Cartie D.D.R. & P.E. Irving. (2002) on
pure Carbon.
The matrix crack that formed because of the impact usually propagates and ends at a place
where it meets a stiffer fiber. Since the adopted stacking sequence has stiffer fibers covering
all the directions, the crack propagation was terminated. In comparison with the damaged
Figure 2. Impacted sample at energy 25J.
359

samples in the literature, it is evident that the adopted method of stacking sequence was
effective using the plain weave hybrids.
Hybridization and stacking sequence tailored the brittle nature of carbon. The plain
wave of fibers prohibited the propagation of the crack and minimized the damage area,as
observed in Figure 4.
Figure 5 shows the CAI damage of Glass-Carbon/epoxy laminate for all three energies.
The crack formed by the compressive force propagates in the direction perpendicular to load-
ing. The intensity of the crack and sub laminate buckling increased with the increase of
impact energy.
The Force-Time history of Glass-Carbon/epoxy was given in Figure 6. The maximum
force required for the sample to fail decreased with the increase in impact energy. The time to
failure for the 35J and 45J were near. However, the impact of 25J energy took a longer time
to failure.
Figure 7, shows the Force vs Displacement curve for all the three energies. The curves were
linear and hence the displacement decreased with the increase of impact energy. The Dis-
placement vs Time curve was exactly linear and proportional to each other for all the ener-
gies as shown in Figure 8. However, the time to failure and the displacements were inversely
proportional. Higher energies require a lesser time and lesser energy requires a longer time.
Figure 5. CAI crack formation for (a) 25J (b) 35J (c) 45J.
Figure 6. CAI result of Force vs Time.
Figure 7. CAI result of Force vs Displacement.
360

Figure 8. CAI result of Displacement vs Time.
Figure 9. SEM images of CAI at energy 25J.
Figures 9–11 shows the SEM image of CAI specimen for energies 25J, 35J and 45J respec-
tively. From the figures, it is evident that failure occurred because of the sub laminate buckling.
Matrix crack was dominant for the energy at 25J. The dominant parameters for the laminates
of energy greater than 25J were fiber pullout, fiber micro cracking and Delamination.
The damage of the CAI happens in the direction perpendicular to loading. The tensile or
the shear load that occurs as a result of impact initiates the matrix crack, which propagates
through the thickness of the laminate.
Figures 10 and 11 shows the SEM images of energy greater than 25J. These images witness
the fiber pullout, fiber fracture, matrix fiber de-bonding and results in severe plastic deforma-
361

362

tion. The larger value of in-plane properties of Carbon and the higher impact property of
Glass complimented each other to result in an increased performance.
Dent in the matrix caused by the fiber pull out is clearly visible in Figure 11. Micro cracking
of the fiber is visible only in carbon because of its brittle nature.
Quasi isotropic stacking sequence with plane weave fabrics has an added advantage espe-
cially to the impact response. Stacking sequence is also a main parameter to determine the
CAI strength of the laminate.
4 CONCLUSION
From the above experimentation, the authors found out the following:
• Crack propagation due to impact can be terminated by stiffer fibers in the path of crack.
• For energies less than 25J, matrix crack is predominant.
• For energies greater than 25J other parameters such as fiber pullout, fiber micro cracking
and Delamination.
• Delamination occurs as a result of sub laminate buckling.
• The contact force and time were inversely proportional.
• The time to failure and the displacement were inversely proportional.
• Stacking sequence is also a main parameter to determine the CAI strength of the
laminate.
• Hybridization results in an increased performance, where two or three fibers in a stacking
sequence, as mentioned in this research, can complement each other in the performance as
a single laminate.
REFERENCES
Alaattin Aktas et al. 2014. Impact and post impact (CAI) behavior of stitched woven-knit hybrid com-
posites.Composite Structures. 116, 243–253.
AndreyShipsha & Dan Zenkert. 2005. Compression after impact strength of sandwich panels with core
crushing damage. Applied Composite Materials. 12, 149–164.
Aymerich F. & P. Priolo. 2008. Characterization of fracture modes in stitched and unstitched cross-
ply laminates subjected to low-velocity impact and compression after impact loading. Int. J. Impact
Engineering. 35, 591–608.
Berketis K. & D. Tzetzis. 2010. The compression after impact strength of woven and non-crimp fabric
reinforced composites subjected to long term water immersion ageing. J. Mater Sci. 45, 5611–5623.
Bin Yang et al. 2015. Study on the low velocity impact response and CAI behavior of foam filled sand-
wich panels with hybrid facesheet. Composite Structures. 132, 1129–1140.
Bruno Castanie et al. 2008. Core crushing criterion to determine the strength of sandwich composite
structures subjected to compression after impact, Composite Structures, 86, 243–250.
Cao D.F. et al. 2015. Compressive properties of SiC particle reinforced aluminum matrix composites
under repeated impact loading. Strength of Materials. 47, 61–67.
Cartie D.D.R. & P.E. Irving. 2002. Effect of resin and fibre properties on impact performance of CFRP.
Composites: Part A. 33, 483–493.
Daniele Ghelli & Giangiacomo Minak. 2011. Low velocity impact and compression after impact tests
on thin carbon/epoxy laminates. Composites: Part B. 42, 2067–2079.
Davies G.A.O. et al. 2004. Compression after impact strength of composite sandwich panels. Composite
Structures. 63, 1–9.
Gilioli A. et al. 2014. Compression after impact test (CAI) on NOMEX honeycomb sandwich panels
with thin aluminum skins. Composites: Part B. 67, 313–325.
Gonzalez E.V. et al. 2012. Simulation of drop weight impact and compression after impact tests on
composite laminates. Composite Structures. 94, 3364–3378.
Guoqi Zhang et al. 2013. The residual compressive strength of impact damaged sandwich structures
with pyramidal truss cores. Composite Structures. 105, 188–198.
Hakim Abdulhamid et al. 2016. Experimental study of compression after impact of asymmetrically
tapered composite laminate. Composite Structures. 149, 292–303.
363

Hakim Abdulhamid et al. 2016. Numerical simulation of impact and compression after impact of asym-
metrically tapered laminated CFRP. International J. Impact Engineering. 95, 154–164.
Hao Yan et al. 2010. Compression-after-impact response of woven fiber-reinforced composites. Com-
posite Science and Technology. 70, 2128–2136.
Hao-Ming Hasiao. 2012. Compression after impact strength and surface morphology in toughned com-
posite materials. Int J. Fract. 176, 229–236.
Jefferson Andrew J. et al. 2015. Compression after impact strength of repaired GFRP composite lami-
nates under repeated impact loading. Composite Structures. 133, 911–920.
Jun-Jiang Xiong et al. 2008. A strain based residual strength model of Carbon/epoxy composites based
on CAI and fatigue residual strength concepts. Composite Structures. 85, 29–42.
Kostopoulos V. et al. 2010. Impact and compression after impact properties of carbon fibre reinforced
composites enhanced with multi wall carbon nanotubes. Composites science and technology. 70,
553–563.
Mandar D. Kulkarni et al. 2011. Effect of back pressure on impact and compression after impact char-
acteristics of composites. Composite Structures. 93, 944–951.
Mannov E. 2013. Improvement of compressive strength after impact in fibre reinforced polymer com-
posites by matrix modification with thermally reduced graphene oxide. Composite Science and Tech-
nology. 87, 36–41.
Mehmet Aktas et al. 2009. Compression after impact behavior of laminated composite plates subjected
to low velocity impact in high temperatures. Composite Structures. 89, 77–82.
Mehmet Aktas et al. 2012. Impact and post impact behaviour of layered fabric composites. Composite
Structures. 94, 2809–2818.
Michael Dale et al. 2012. Low velocity impact and compression after impact characterization of woven
carbon/vinylester at dry and water saturated conditions. Composite Structures. 94, 1582–1589.
Michael et al. 2010. Compression after impact of sandwich composite structures: Experiments and
Modelling. AIAA. 2867, 1–16.
Panettieri E. et al. 2016. Delaminations growth in compression after impact test simulations: Influence
of cohesive elements parameters on numerical results. Composite Structures. 137, 140–147.
Samuel Rivallant et al. 2013. Failure analysis of CFRP laminates subjected to compression after impact:
FE simulation using discrete interface elements. Composites: Part A. 55, 83–93.
Sanchez Saez S. et al. 2005. Compression after impact of thin composite laminates. Composites Science
and Technology. 65, 1911–1919.
Slattery, P.G. et al. 2016. Assessment of residual strength of repaired solid laminate composite materials
through testing, Composite Structures. 147, 122–130.
Tan K.T. et al. 2012. Effect of stitch density and stitch thread thickness on compression after impact
strength and response of stitched composites. Composites Science and Technology. 72, 587–598.
Tan K.T. et al. 2015. Finite element model for compression after impact behavior of stitched compos-
ites. Composites: Part B. 79, 53–60.
Thomas E. Lacy & Youngkeun Hwang. 2003. Numerical simulation of impact damaged sandwich com-
posites subjected to compression after impact loading.Composite Structures. 61, 115–128.
Vaidya U.K. et al. (2000). Low velocity impact and compression after impact response of Z-pin rein-
forced core sandwich composites. Transactions of the ASME. 122, 434–442.
Wei Tan et al. (2015). Prediction of low velocity impact damage and compression after impact (CAI)
behavior of composite laminates. Composites: Part A. 71, 212–226.
Xinguang Xu et al. (2014). Improving compression after impact performance of carbon fiber compos-
ites by CNTs/thermoplastic hybrid film interlayer. Composites science and technology. 95, 75–81.
Yuichiro et al. (2008). Effect of hygrothermal condition on compression after impact strength of CFRP
laminates. Composites science and technology. 68, 1376–1383.
Zhang X et al. (2006). Improvement of low velocity impact and compression after impact performance
by z-fibre pinning.Composites Science and Technology. 66, 2785–2794.
364

Effect of polarity in micro-electrical discharge machining
Jibin Boban, Arun Lawrence & K.K. Manesh

Department of Mechanical Engineering, Government Engineering College, Thrissur, India
Leeba Varghese
Department of Mechanical Engineering, Viswajyothi College of Engineering and Technology,
Ernakulam, India
ABSTRACT: Micro Electric Discharge machining can be used to generate micro features
and micro level dimensions on the work-piece irrespective of the hardness of the material.
This paper discusses the effect of polarity in tool wear during micro-EDM drilling of stain-
less steel work-piece (SS 304). An experimental investigation has been carried out to under-
stand the effect of change in polarity in tool wear using three different tool electrodes (Cu,
Brass and W). Direct polarity has significant impact over reverse polarity in reducing tool
wear for all the three electrodes. Further, observations indicated that material removal rate for
stainless steel is maximum in case of direct polarity.
Keywords: Tool wear rate, Tool electrodes, Material removal rate, Polarity
1 INTRODUCTION
EDM is a non-traditional machining process which involves the removal of electrical conduc-
tive material by a series of electric sparks between two electrodes submerged in a dielectric
fluid. The material removal mechanism involves the melting and vaporization of the work-
piece material caused by these electric sparks.
In the current scenario, micromachining of materials has become essential to make pre-
cise and accurate components (Yuangang et al. 2009). Micro-EDM is a recently developed
method that can be used for producing micro-parts within the range of 50 µm–100 µm. It is
an efficient machining process for the fabrication of miniaturized products, micro channels,
micro-metal holes and micromold cavities with a lot of merits resulting from its character-
istics of non-contact and thermal metal removal process (Yeakub Ali & Mohammed 2009).
The tool wear in micro-EDM directly affects the machining precision and efficiency
(Jingyu et al. 2017). Hence the minimization of tool wear is of great importance in micro-edm
Figure 1. Schematic diagram for micro-EDM (Equbal et al. 2009).
365

process to achieve the required accuracy of machined features (Bissacco et al. 2010). In macro-
EDM, the workpiece is made positive (anode) and the tool electrode is made the cathode. This
is referred to as direct or straight polarity (Cyril et al. 2017). Similarly, if the tool electrode
is made positive, it is referred to as negative polarity. The influence of change in polarity in
micro-EDM is less studied by researchers. Thus, experiments are conducted by changing the
polarity of tool electrode or work-piece in order to study its effect on tool wear in micro-EDM.
In the study of Yoshiyuki et al. (1991), change in polarity has significant effect on the electrical
discharge machining performance. Also, Lee & Li (2001) have reported that negative polarity
of tool offers less tool wear.
In micro-EDM, the machinability of the work-piece material mainly depends on the ther-
mal conductivity and melting point (Yu et al. 2014). Stainless steel (SS 304) has high melting
point in the range of about 1400–1455°C. But the tool electrodes used for machining have
melting point lower than the work-piece material. So chance for tool wear is high which
makes the study important. Experimental investigation is carried out to determine how the
change in polarity affects the tool-wear. Also the effect of polarity in the material removal
rate of stainless steel is also checked for both the polarities.
2 EXPERIMENTAL SETUP DETAILS
An In House built Micro-EDM was used for the experimental investigation. Stainless steel speci-
mens (30 × 20 × 5 mm) for experiment were cut using abrasive cutters. Copper, Brass & Tungsten
are the tool electrodes used for study. The details of the experiment are given in Table 1:
The major input parameters used in micro-EDM are Input voltage, Input current, Pulse
On time and Pulse off-time. By conducting a lot of pilot experiments, optimum values of
process parameters which gives good machining is identified and selected for the study. The
experiment levels are presented in Table 2:
The properties of the tool electrodes and work-piece used are given in Tables 3 and 4.
Experiment is carried out using separate tools and work-pieces for both direct and reverse
polarities. The weights of both tool and work-piece are noted using precision weighing bal-
ance, before and after machining in each case.
Material Removal Rate (MRR) and Tool Wear Rate (TWR) are calculated using the
equation:
Weight before machining − Weight after machining

MRR =
Machini
ing Time
Weight before machining − Weight after machining
TWR =
Machini
ing Time
Table 1. Experimental details.
Work-Piece Stainless Steel (SS 304) 30 mm × 20 mm × 5 mm
Tool Cu, Brass, W 0.8 mm φ, 40 mm length

Dielectric De-ionized water –
Table 2. Experimental conditions.
SI No Parameters Values
1 Input Voltage 50 V
2 Pulse On-time 110 μS
3 Pulse Off-time 30 μS
366

Figure 2. In-house built micro-EDM experimental setup.
Table 3. Properties of tool electrodes.
Specifications of electrodes
Properties Copper Brass Tungsten
Density (kg/m3) 8910 8490 19.25

Thermal Conductivity (W/m-K) 392 158 173
Melting Point (°C) 1083 900 3422
Table 4. Properties of stainless steel (SS 304).
Properties Values
Brinell Hardness 123

Density 8000 kg/m3
Thermal Conductivity 16.2 W/m-K
Melting Point 1400–1455°C
The effect of change in polarity on the tool wear rate and Material removal rate are discussed.
The results are based on the experimental investigation performed by machining stainless
steel (SS 304) specimen.
The response table for TWR and MRR is shown in Table 5:
3.1 Effect of polarity on tool wear

Figure 3 indicates the effect of change in polarity on tool wear rate. It can be noticed that the
tool wear is minimum in straight or direct polarity compared to reverse polarity for all the three
electrodes. In positive polarity, the tool is made the cathode and work piece is made the anode.
Electrons are lighter in mass and hence they get accelerated faster from the cathode (tool elec-
trode). Therefore electrons bombarding on the anode (work-piece material) will generate more
heat energy than positively charged particles hitting the tool electrode. This helps to reduce the
367

Table 5. TWR and MRR for different tool electrodes.
TWR (g/min) MRR (g/min)
Tool electrode Direct Reverse Direct Reverse
Copper 0.0028 0.0149 0.0036 0.0024

Brass 0.0032 0.0135 0.0028 0.001
Tungsten 0.0007 0.0017 0.0025 0.0012
Figure 3. Effect of polarity on tool wear rate.
tool wear rate. The tool wear is reduced by about 81.2% for copper tool, 76.3% for brass tool
and 58.8% for Tungsten tool when direct polarity is used instead of reverse polarity.
3.2 Effect of polarity on MRR

Figure 4 indicates the effect of change in polarity on material removal rate for stainless steel. It
can be noticed that the MRR is minimum in reverse polarity compared to direct polarity for all
the three electrodes. This is due to the high kinetic energy of electrons as compared to that of a
positively charged ion. The high motion energy of electrons may be attributed to the light weight
of electron and the high acceleration of the electric field. Thus, electron bombards the anodic
surface and gives more energy to the work-piece, thereby increasing the material removal rate.
3.3 Comparison of tool electrodes

The three tool electrodes show different tool wear characteristics during micro electrical dis-
charge machining of the work-piece. From Figure 3, it can be observed that, in both polari-
ties, tool wear is minimum for Tungsten electrode. This is due to the high melting point of
tungsten electrode (3422°C) compared to that of stainless steel work-piece (1450°C).Also
maximum MRR is obtained by copper tool electrode irrespective of the polarities. This can
be inferred from Figure 4.
368

3.4 Physical evaluation of work-piece surface
It is also observed from visual inspection that large deposition of tool material on the work-
piece surface occurs in reverse polarity compared to that of direct polarity. Figure 5 shows the
images of work-piece surface obtained by means of image acquisition system. The amount
of tool deposition has to be further analyzed using spectrometric analysis or using Scanning
Electron Microscopy.
Figure 4. Effect of polarity on material removal rate.
Figure 5. Tool deposition on the machined surface for three electrodes.
369

4 CONCLUSIONS
In this study, the effect of change in polarity is evaluated for various tool electrodes by
measuring TWR and MRR. The following conclusions are derived from the experimental
investigation:
1. In direct polarity, tool wear is found to be minimum compared to that of reverse polar-
ity on account of more heat generation at the work-piece surface as in case of direct
polarity.
2. Direct polarity also offers large MRR than reverse polarity for all the three electrodes.
3. The comparative study of Cu, Brass and W tool electrodes revealed that Tungsten tool
offers minimum tool wear in both direct and reverse polarities.
4. High MRR is provided by the copper tool electrode while machining stainless steel.
5. From physical observation, deposit of tool electrode material on the work-piece surface is
higher in case of reverse polarity.
The study can be further extended to determine the effect of change in polarity on other
work-piece materials rather than stainless steel to identify the suitable polarity that can be
used in micro-EDM. In the present paper, study on tool electrode wear is given more focus
and hence surface roughness is not analyzed. Since surface roughness is too a major output
parameter in micro-EDM, further studies are recommended to study the effect of polarity
on surface roughness.
ACKNOWLEDGEMENT
The authors would like to acknowledge the financial support of the Centre for Engineering
Research and Development (Proceedings no. C3/RSM86/2013, dated 09/02/2015 of Kerala
Technological University), Government of Kerala, India.
REFERENCES
Bissacco, G. Valentincic, J. Hansen H.N. & Wiwe. B.D. 2010. Towards the effective tool wear conrol in
micro-EDM milling. Int. J. Adv. Manuf. Technol., 47: 3–9.
Cyril Pilligrin, J., Asokan, P. Jerald, J. Kanagaraj G., J.M. Nilakantan & Nielsen, I. 2017. Tool speed and
polarity effects in micro EDM drilling of 316 L stainless steel. Production & Manufacturing Research,
5(1): 99–117.
Equbal, A. & Sood, A.K. 2014. Electric Discharge Machining; An overview on various areas of
research. Journal of Manufacturing and Industrial Engineering, 13: 1–6.
Jingyu, P., Zhang, L., Du, J., Zhuang, X., Zhou, Z., W. Shunkun & Zhu, Y. 2017. A model of tool wear
in electrical discharge machining process based on electromagnetic theory. International Journal of
Machine Tools & Manufacture, 117: 31–41.
Lee, S.H. & Li, X.P. 2001. Study of the effect of machining parameters on the machining characteristics
in electrical discharge machining of tungsten carbide, Journal of Materials Processing Technology,
115: 344–358.
Liua, Y., Zhangb, W., Zhangc, S. & Sha, Z. 2014. The Simulation Research of Tool Wear in Small Hole
EDM Machining on Titanium Aloy, Applied Mechanics and Materials, 624:249–254.
Yeakub Ali, M. & Mohammad, A.S. 2009. Effect of Conventional EDM Parameters on the Micro
machined Surface Roughness and Fabrication of a Hot Embossing Master Micro tool, Materials and
Manufacturing Processes, 24: 454–458.
Yuangang, W. Fuling, Z. & Jin, W. 2009. Wear-resist Electrodes for Micro-EDM, Chinese Journal of
Aeronautics, 22: 339–342.
370

Characterization and Taguchi based modeling and analysis of

dissimilar TIG welded AISI 316L austenitic stainless steel-HSLA
steel joints
P.V. Shaheer, Anwar Sadique & K.K. Ramachandran

ABSTRACT: In this study, 3 mm thick plates of AISI 316L austenitic stainless steel and
High Strength Low Alloy (HSLA) steel were dissimilar GTA welded and the tensile strength
and microstructural properties were investigated. The experimental trials were carried out
as per the Taguchi design and the welding current, welding speed, wire feed rate and filler
material were selected as the parameters. The results showed that the highest joint strength of
610 MPa was obtained at welding current of 100 A, welding speed of 9 cm/min and wire feed
rate of 1.6 m/min with 304 steel as the filler material. ANOVA revealed that the welding cur-
rent and wire feed rate are the most significant and least significant parameters, respectively.
The regression model showed that the welding speed and filler material and the welding cur-
rent and wire feed rate have interaction effect on the tensile strength of the joints.
1 INTRODUCTION
Austenitic stainless steel, AISI316L is known for its inherent superior properties like high
strength at elevated temperatures, increased resistance to pitting and general corrosion, high
creep strength etc. High strength low alloy (HSLA) steels possess very high strength, high
strength to weight ratio and increased corrosion resistance together with relatively low cost.
316L austenitic stainless steel and HSLA steels have wide range of combined application in
the marine, automotive and locomotive sectors. In spite of having excellent qualities, indi-
vidually, dissimilar fusion welded 316L/HSLA steel joints (by conventional techniques) are
prone to complications such as grain coarsening, sensitization, stress corrosion cracking, hot
cracking etc. Thus, accurate control of welding process parameters is essential to regulate the
heat input and hence to reduce the associated problems.
With regard to fusion welding of austenitic stainless steels, Anand Rao V & Daivanathan
B (2015) conducted detailed experimental study on TIG welding of 310 steel joints. In their
research, a total of 9 welded joints were fabricated and tested with the objectives of analysis
and optimization of the TIG welding process. The results showed that welding current of
120 A with 309L steel filler metal had produced the highest joint tensile strength of 454.6 MPa
while a welding current of 80 A with 316L filler metal produced the lowest tensile strength
of 517.9 MPa. Navid Moslemi et al. (2009) conducted a study on the effect of welding cur-
rent on the mechanical and microstructural characteristic of TIG welded 316 steel joints. The
mechanical characteristics of the welded joints such as tensile strength and microhardness
were evaluated in the study. Microstructural studies confirmed the presence of secondary
sigma phase that caused embrittlement in the weld zone.
Bharatha et al. (2013) reported the process optimization and joint analysis of 316 steel
TIG welded joints using Taguchi technique. With regard to dissimilar fusion welding of
stainless steels and HSLA steels, only a very few works are reported in the literature. Anant
et al. (2017) have developed a special nozzle for the GMAW and successfully welded 25 mm
thick plates of dissimilar 304L stainless steel and SA543 HSLA steel. The authors have used
308 L steel as filler metal and the weld was completed by multiple passes.
371

To the best of knowledge of the authors, studies on dissimilar fusion welding of 316L and
HSLA steels are not reported in the literature. Also, implementation of the dissimilar joining
techniques needs the knowledge of the effect of operating parameters on the performance of
the joints and optimization of the welding process. Therefore, in this work, austenitic stain-
less 316L and HSLA steel are dissimilar TIG welded and the influence of process parameters
on the joint performance are investigated.
The base materials selected for this research were 3 mm thick rolled sheets of HSLA steel,
IRS-M42-93 and austenitic stainless steel, AISI 316L. The composition of HSLA steel
and 316L steel is given in Table 1. The optical micrographs of the base metal are shown in
Figure 1. The base 316L steel consists of mostly coarse austenite with minor amount of fer-
rite. Annealed twins crossing the grain boundary interface were observed. The base HSLA
steel is with approximately equiaxed and fine grained ferritic–pearlitic microstructure. Work
pieces of 150 mm × 50 mm × 3 mm size were cut by cold shearing. Three types of filler wires
were used in this investigation; AISI 304 steel, 309L steel and 316L steel. The composition of
each of the filler material is shown in Table 3. The welding trials were performed using auto-
mated TIG welding equipment and the joints were finished in a single pass welding.
The Taguchi – L9 orthogonal design is selected as the design of experiment (DOE) tech-
nique (Krishnaiah, K. & Shahabudeen, P. 2012). The most important stage in the DOE is the
very selection of DOE technique and the selection of the control factors. The travel speed
(welding speed), welding current, wire feed rate and filler material were selected as the factors
(variables) of the design. The range and upper and lower bounds of the selected variables
were fixed by trial and error with the criterion of visually defect free joints. The four factor,
three level; L-9 design matrix given in Table 2 was developed using the statistical software,
Minitab 17. Two sets of joints were fabricated at each set of factor setting and the average of
the response was considered for modeling and analysis.
Table 1. Chemical composition of the base materials.
Composition (wt%)
Material C Mn Si P S Al Cu Cr Mo Ni Nb Ti V N Fe
HSLA Steel; 0.11 0.42 0.32 0.10 0.01 0.029 0.31 0.54 0.001 0.22 0.001 0.002 0.002 – Bal
IRSM-42-97
AISI 316L steel 0.03 2.0 0.75 0.045 0.03 – – 17 2–3 10–14 – – – 0.1 Bal
Figure 1. Optical micrographs of base materials; (a) 316L steel (b) HSLA steel.
372

Table 2. The L-9 Taguchi DOE matrix.
Sl. No. Filler material Current (A) Travel speed (cm/min) Feed (m/min)
1 SS 304 80 8 1.2
2 SS 304 90 8.5 1.4
3 SS 304 100 9 1.6
4 SS 316L 80 8.5 1.6
5 SS 316L 90 9 1.2
6 SS 316L 100 8 1.4
7 SS 309L 80 9 1.4
8 SS 309L 90 8 1.6
9 SS 309L 100 8.5 1.2
Table 3. Chemical composition of the filler materials materials.
Composition (Wt. %)
Material Ni Cr Si Mn C P S Mo N Cu Fe
AISI 304 9.25 19 1 2 0.08 0.045 0.03 – – – Bal

AISI 316L 10–14 17 0.75 2 0.03 0.045 0.03 2–3 0.10 – Bal
AISI 309L 12–14 23.5 0.3–.65 1–.5 0.03 0.03 0.03 0.75 – 0.75 Bal
The specimens for the tensile test were cut by conventional milling process with the geom-
etry as per ASME SEC IX (2015): Boiler and Pressure Vessel Code (QW-462 Test specimen).
The tensile test was done on a universal testing machine with 50 kN capacity at crosshead
speed of 4 mm/min. The specimen for optical microscopy was prepared by using standard
metallographic procedures. The etched specimens were observed under optical microscope
model: BX51.
Visual examination of welded joints showed that in seven out of nine welding conditions,
no visible surface defects were observed. But, for the sixth and seventh runs, the joints were
observed with visual defects such as excessive deposition and lack of fusion on the HSLA
side, respectively. The observed surface defects may be caused due to excessive/low heat gen-
eration as a result of the combined effect of welding current, welding speed, and filler metal
feed rate.
The average tensile strength of the joints is shown in Table 4. The highest tensile strength
of 610 MPa and the least strength of 415 MPa is obtained for the fifth and third trail run,
respectively. The third trial that resulted the highest joint strength is corresponding to weld-
ing current of 100 A, welding speed of 9 cm/min and feed rate of 1.6 m/min. Whereas the
joint with the least strength correspond to welding current of 80 A, welding speed of 8.5 cm/
min and wire feed rate of 1.6 m/min. For the best joint (highest strength), the filler material
used was 304 steel whereas for the joint with the least strength, the filler material was 316L
steel.
Figures 2 and 3 portray typical optical micrographs of the weld nugget region of the joints
that possess the lowest and highest joint strength, respectively. Figure 2 reveals the formation
of coarse grain in the nugget zone. Also, it seems that the grain boundaries are characterized
by the precipitation of secondary phase. The precipitation may be sigma phase as a result of
the very high heat generation and the use of 316L filler metal during welding. Under high
heat input, there will be increased distribution of sigma phase in nugget zone with a wider
373

Table 4. Coded and actual values of the variables and values of tensile strength.
Coded values of
variables Actual values of variables
Sl. M I U F TS S/N
No. M I U F (Grade) (A) (cm/min) (m/min) (Mpa) Ratio
1 1 1 1 1 304 80 8 1.2 420 52.4650

2 1 2 2 2 304 90 8.5 1.4 549 54.7914
3 1 3 3 3 304 100 9 1.6 610 55.7066
4 2 1 2 3 316L 80 8.5 1.6 415 52.3610
5 2 2 3 1 316L 90 9 1.2 596 55.5049
6 2 3 1 2 316L 100 8 1.4 471 53.4604
7 3 1 3 2 309L 80 9 1.4 530 54.4855
8 3 2 1 3 309L 90 8 1.6 533 54.5345
9 3 3 2 1 309L 100 8.5 1.2 459 53.2363
Figure 2. Optical micrographs of the weld nugget of the lowest strength joint.
Figure 3. Optical micrographs of the weld nugget of the highest strength joint.
heat affected zone (HAZ). Under tensile load, cracks can easily propagate through the grain
boundaries causing failure at low loads. Thus the observed microstructure clearly substanti-
ates the lowest joint strength resulted for the joint.
Referring to Figure 3, it seems that the weld nugget is with low concentration of second-
ary phase formations. The marginally low heat generation and the use of 304 steel filler wire
would be the probable causes for the low concentration of secondary phases. Vitek, J. M. &
David, S. A. (1984) reported that the sigma phase reaction is accelerated by the large scale
374

atomic movement (advance of re-crystallization front) rather than the chromium content
in steel. Therefore, though 304 steel is having higher chromium content, the higher welding
speed could be the probable reason for the higher joint tensile strength.
3 MODELING AND ANALYSIS
The response, tensile strength (TS) of the joint is modeled as a function of the four factors
that were selected for the DOE. The generalized form of the regression model is given by;
TS = f (I, U, F, M) (1)
where I is the welding current, U is the welding speed, F is the wire feed rate and M is the
filler material. The coded and actual values of the variables and the values of TS are given in
Table 4. The regression model developed using the proprietary statistical software; Minitab17
is given in equation (2).
TS = 248.1 − (28.7 × M)+ (283.7 × I)+

(7.3 × U)− (77 × F) − (86.33 × 2 (2)
× I )+
(22.33 × M × U)+ (51 × I × F)
The adequacy of the developed model is verified by the analysis of variance (ANOVA) and
the actual and adjusted coefficients of determination (R2). The results of the ANOVA are
shown in Table 5. The larger values of F-ratio indicate that the model is adequate and capable
to predict the TS accurately. The actual and adjusted R2 values of the model are 95.96% and
99.65%, respectively. Considering the nature of variability of the welding process, the above
values of R2 shows adequacy of the developed model. Conformity experimental trials agree
well with the TS values predicted by the model with an average error of 0.82%.
3.1 Analysis of the developed model

From equation (2) it can be seen that only the factors, current and wire feed rate and filler
material and welding speed have mutual interaction effect on the TS of the joints. The devel-
oped model is analyzed to illustrate the individual and combined effects of the factors (param-
eters) on the response (TS) of the joints. The individual and combined effects of parameters
on TS are plotted by keeping the other parameters at the middle values.
Referring to Figure 4 (a), it can be seen that for the mid values of current and wire feed rate
(90 A and 1.4 m/min), at welding speeds of 8.5 and 9 cm/min, 309L filler material is giving
the highest joint strength. But, at low welding speed of 8 cm/min, the best filler material is
304 steel. Also, at low heat input, the suitable filler materials are 316L and 309L. This may be
Table 5. ANOVA of the developed model.
Source Adjusted sum of squares Adjusted mean squares F-Value P-Value
Regression 41069.8 5867.11 330.03 0.042

M 99.4 99.41 5.59 0.255
I 3613.8 3613.78 203.27 0.045
U 6.6 6.59 0.37 0.652
F 2615.7 2615.74 147.14 0.052
I2 7453.4 7453.44 419.26 0.031
M*U 249.4 249.39 14.03 0.166
I*F 6502.5 6502.50 365.77 0.033
Error 17.8 17.78 – –
Total 41087.6 – – –
375

due to the evolution of approximately equal amount of austenitic and ferritic phases in the
weld nugget, rather than the formation of chromium carbide.
Figure 4(b) show the influence of welding current on the tensile strength of the joints at
different wire feed rates. The plots clearly show that the welding current and wire feed rate
have significant interaction on the joint strength. Also, the highest and lowest joint strength
is resulted for wire feed rate of 1.6 m/min. The peak joint strength corresponds to 95 A and
1.6 m/min. The proper material deposition under moderate heat input and wire feed rate
and the resulted better microstructure could be the probable reason for the higher joint
strength.
Figure 4. Effect of welding parameters on joint strength.
Figure 5. Contour plot of Tensile Strength (TS) vs current & feed rate and TS vs filler metal &
welding speed.
376

The influence of wire feed rate on joint strength at different welding currents is shown
in Figure 4(c). Consistently higher joint strength is resulting at welding current of 90 A. At
higher current of 100 A, though the strength is very low at low feed rate, it sharply increases
and reaches the peak value at 1.6 m/min. This indicates that the higher heat input at 100 A is
compensated by the increase wire feed rate and probably resulting better microstructure with
less secondary phase precipitation. Figure 4(d) portrays the influence of welding speed on
joint strength. The joint strength shows an increasing trend as the welding speed increases, at
all the currents. But, the highest strength is obtained at 90 A, welding current. This implies
that the heat input corresponding to 90 A is near optimal and results better nugget micro-
structure with less precipitation of secondary phases. Figure 5 shows the contour plot of ten-
sile strength versus current and feed rate and TS vs filler material and welding speed. Other
parameters do not have any significant interaction effect on the joint strength.
4 CONCLUSION
In the present research, dissimilar AISI316L stainless steel and HSLA steel were successfully
TIG welded and the joint tensile strength and microstructure were investigated. The follow-
ing conclusions were made.
– The highest joint strength of 610 MPa is obtained for the joint fabricated at welding cur-
rent of 100 A, welding speed of 9 cm/min and wire feed rate of 1.6 m/min with 304 steel as
filler metal.
– The micrographs revealed signs of secondary phase evolution at the grain boundaries in
the weld nugget region.
– ANOVA of the results showed that the welding current and wire feed rate are the most
significant and least significant parameter, respectively.
– Analysis of the developed model suggest that the welding current and wire feed rate have
significant interaction effect on the joint strength and the welding speed and filler metal
have moderate interaction effect on the joint strength.
REFERENCES
Anand Rao, V. & Deivanathan, R. 2014. Experimental Investigation for welding aspects of stainless
steel 310 for the Process of TIG welding. Procedia Engineering, 97, 902–908.
Bharath, P. Sridhar, V.G. & Senthil Kumar, M. 2014. Optimization of 316 stainless steel weld joint char-
acteristics using Taguchi technique, Procedia Engineering, 97, 881–891.
Krishnaiah, K. & Shahabudeen, P. 2012. Applied Design of Experiments and Taguchi Methods.
Nabendu Ghosh, Pradip Kumar Pal. & Goutam Nandi 2016. Parametric optimization of MIG welding
on 316L austenitic stainless steel by Grey-based Taguchi method, Procedia Technology 25, 1038–1048.
Navid Moslemi, Norizah Redzuan, Norhayati Ahmad & Tang n Hor. (2015) Effect of current on char-
acteristic for 316 stainless steel welded joint including microstructure and mechanical properties,
Procedia CIRP. 26. 560–564.
Ramachandran, K.K. Murugan, N. & Shashi Kumar, S. 2015. Influence of tool traverse speed on the
characteristics of dissimilar friction stir welded aluminium alloy, AA5052 and HSLA steel joints.
Archives of civil and mechanical Engineering, 15, 822–830.
Ramkishor Anant & Ghosh, P.K. 2017. Ultra-narrow gap welding of thick section of austenitic stain-
less steel to HSLA steel, Journal of Materials Processing Technology, 239, 210–221.
Vitek, J.M. & David, S.A. 1984. The sigma phase transformation in austenitic stainless steels, 65th
Annual AWS Convention in Dallas, Tex.
377

Mechanical design, vibration and tribology

Behavior of stress intensity factor of semi-elliptical crack

at different orientations subjected to thermal load
M.B. Kumaraswamy & J. Sharana Basavaraja

Department of Mechanical Engineering, BMS College of Engineering, Bengaluru, Karnataka, India
ABSTRACT: Fracture mechanics is a very important tool used for improving the life
cycle of mechanical components. During manufacturing, flaws or cracks will be formed in
all metal structures. Studying and monitoring the propagation of the crack in a component
forms the core of fracture study. In the present work, a parametric study on the propaga-
tion of semi elliptical crack in a turbine blade is carried out. A turbine blade with its hub is
modeled using CATIA software. The model is then imported into ANSYS Workbench and
a Finite Element analysis is performed. Rotational velocity is applied on cracks at different
orientations ranging from 0° to 90° at different crack depth to half crack length ratios and
stress intensity factor (K) is determined for two cases with thermal load and without thermal
load i.e. static load case. The Finite Element results are validated by an empirical solution of
Raju-Newman solution using the MATLAB software. The results will be useful in the assess-
ment of structural integrity of the component.
1 INTRODUCTION
Fracture mechanics analysis forms the basis of damage tolerant design methodology. Its
objectives are the determination of stress intensity factor (K), energy release rate (G), path
independent integral (J), Crack Tip Opening Displacement (CTOD) and prediction of mixed
mode fracture, residual strength and crack growth life.
Solution for Stress Intensity Factor (SIF) in mode I for a surface crack in a plate is pre-
sented in empirical form by Newman & Raju (1981). This empirical equation is presented as
a function of parametric angle, depth and length of the crack as well as thickness and width
of the plate for tension and bending loads. Witek (2011) discusses the failure of compressor
blade due to due to bending fatigue loads, and the calculation of SIF is performed using
the Raju-Newman solution for a semi-elliptical crack. Barlow & Chandra (2005) discuss
the fatigue crack growth rates at the fan blade attachment in an aircraft engine due to cen-
trifugal and aerodynamic loads. Song et al. (2007) deliberate on the failure of a jet-engine
turbine blade due to improper manufacturing techniques. One of the observations that can
be made from the above-mentioned discussions is that the cracks may originate in any form
and direction due to improper design or manufacturing methods. Hence it becomes impera-
tive to perform analysis to obtain SIF values of cracks at various orientations, for growth
analysis.
In this work, fracture analysis is performed on the turbine blade of third stage turbine
bucket of a gas turbine. Initially, static analysis is carried out to obtain the region of crack
nucleation, and then semi-elliptical cracks at various orientations with respect to the rotor
axis is analyzed using the finite element technique to obtain the values of stress intensity fac-
tors. A parametric study varying the crack parameters is conducted to analyse their effects on
the stress intensity factor in three modes.
381

2 FINITE ELEMENT MODELLING OF GAS TURBINE BLADES
A turbine blade constitutes a part in the turbine region of a gas turbine engine, and functions
as the component responsible for absorbing energy from the gas at high temperature and
pressure created in the combustor of the engine. The turbine blades very often are the limit-
ing components of gas turbines since they are subjected to extreme thermal and fluid stresses.
Due to this reason, turbine blades are often made out of materials like alloys of Titanium
containing exotic additives. They also use different and ingenious techniques of cooling, such
as air channels inside the blade itself, boundary layer cooling, and thermal barrier coatings.
2.1 Geometric model

Geometric modeling of the turbine blade has been carried out in CATIA. For the present
study, turbine blade of third stage turbine bucket of a gas turbine is considered. To create
the geometric model, aerofoil profile of the blade has been taken from Ref (2005). This data
provides the coordinates of points on the aerofoil that have been imported into CATIA V5
CAD software. The blade profile is attached to a cylindrical hub (Figure 1) with dimensions
ri = 110 mm, ro = 137.5 mm.

Turbine blades operate under very tough conditions of pressures and temperatures, and
hence the blade material should have the capacity to resist failing under corrosion, fatigue
and impact loading. An alloy of Titanium (Ti-6Al-2Sn-4Zr-2Mo) has been selected for the
turbine blade material (2016). Table 1 specify the physical properties of Titanium alloy.
Figure 1. Turbine blade attached to the hub.
Table 1. Physical properties: Titanium alloy (2016).
Physical property Value
Density 4540 kg/m3
Young’s modulus 120 GPa
Shear modulus 45.5 GPa
Poisson’s ratio 0.32
Ultimate tensile strength 1010 MPa
Yield strength 990 MPa
Fracture toughness 148 MPa√m
Coefficient of thermal expansion 8.1 × 10-6/°C
T1 = 500°C
Operating temperatures T2 = 550°C
T3 = 601°C
382

2.3 Validation of finite element model
Finite Element Modelling of the turbine blade has been carried out using ANSYS-Workbench
using 3D tetrahedral elements. In order to validate the FE model, the benchmark solution
from Raju & Newman (1981) has been considered. The geometric dimensions for the bench-
mark plate geometry [9] is as shown in Figure 3: h = 250 mm; b = 250 mm; t = 200 mm; Crack
dimensions: Width 2c = 80 mm; Depth a = 80 mm. The plate and crack dimensions meet the
criteria mentioned for the application of the Raju-Newman solution, which is,
0 < a / c ≤ 1; 0 ≤ a / t < 1; c / b < 0.5

This problem involves a plate with a semi-elliptical crack subjected to tensile load with the
opposite face being fixed. The FE model is shown in Figure 2.
The benchmark solution (1981) is an empirical equation for the stress-intensity factors for
a surface crack. The empirical equation covers a wide range of configuration parameters.
The ratios of crack depth to plate thickness and the ratios of crack depth to crack length
range from 0 to 1. The stress-intensity factor equation (1981) for combined tension and bend-
ing loads is given by,
 a  a a c 
K I = (St + HSb ) Π   F  , , ,φ  (1)
Q  t c b 
For, 0 < a / c ≤ 1; 0 ≤ a / t < 1; c / b < 0.5 and 0 ≤ ∅ ≤ π .

Where Q is an approximation given by,
1.65
a 
Q = 1 + 1.464   (2)
c 
The term HSb can be ignored because only the tensile loading is considered. Hence, the
equation for stress intensity factor becomes
 a  a a c 
K I = (St ) Π   F  , , ,φ  (3)
Q  t c b 
The Raju-Newman solution for the stress intensity factor is for tensile and bending stress.
And since only the rotational loading is considered for analysis, the bending load part of
Figure 2. Finite element model of the plate with a semi-elliptical crack.
383

Figure 3. ANSYS solution vs Raju-Newman solution.
the solution is ignored. Hence, coming to the boundary conditions applied to the plate, the
bottom face of the plate is considered as a fixed support and the on the top face, a tensile
pressure is applied to simulate the centrifugal load. This simulates the crack opening mode.
Figure 3 shows a comparison of mode 1 stress intensity factor values obtained from ANSYS
Workbench and the target solution of Raju-Newman empirical equation given by (3).
The present solution is in good agreement with empirical solution of (1981). The maxi-
mum variation is found to be 2.457%. These values can be considered satisfactory and hence
the Finite Element model developed is considered to be validated.
3 FRACTURE MECHANICS ANALYSIS
3.1 Location of the crack on blade

It is essential to determine the critical location of crack on the turbine blade where the
crack may nucleate. To determine the location, a static analysis is performed on the blade by
applying rotational load on it. By finding the region where the maximum von-Mises stress is
induced, the location of crack can be approximated.
To simulate the centrifugal loading on the turbine blade the central hub is fixed by
constraining all the degrees of freedom of the nodes present in its mesh. A rotational velocity
is imparted at the center of the hub along its axis by creating a new coordinate system.
Static analysis is carried out for 10,000 RPM. The maximum stress is found approximately
at bottom/near root of the turbine blade. Considering the above result, the crack for the
analysis has been introduced in the region of maximum von-Mises stress.
3.2 Finite element model of semi-elliptical crack

The base mesh around the Fracture Affected Zone comprises of tetrahedral elements.
The convergence study was carried out and the final details of the finite element model and
crack mesh is given in Figure 4.
3.3 Evaluation of SIF

Once the meshing is completed as shown in Figure 4 and boundary conditions are applied
and the analysis is performed. Here, the stress intensity factors in mode 1, mode 2 and mode
3 are calculated. Parametric study done with the crack length and depth as the varying para
meters is discussed following paragraphs. Figure 5 shows the plots of stress intensity factor
in mode-1 along the crack front of an example analysis.
384

Figure 4. Blade and crack mesh. Figure 5. SIF plot for opening mode.
Figure 6. K1 for crack angles 0° to 90° at the Figure 7. K2 for crack angles 0° to 90° at the
interval of 15°. interval of 15°.
Figure 8. K3 for crack angles 0° to 90° at the interval of 15°.
The crack mesh has 7 solution contours for each a/c ratio. The Post-Processing of the solu-
tion involves exporting the values to Microsoft Excel for further analysis. For the parametric
studies, the average of the values obtained from all seven solution contours is considered.
Seven different analyses of the blade are carried out with the crack at 0°, 15°, 30°, 45°,
60°, 75° and 90° orientation with respect to the longitudinal axis of the turbine rotor. The
crack length and depth in all the seven cases are kept constant at 2.0 mm and 0.80 mm (2002)
respectively. The plots of SIF along the crack front for the crack angles 0° to 90° at the inter-
val of 15° are shown in Figures 6, 7 and 8.
The behavior of the values of K1 is shown in Figure 6. It is observed that the values of
K1 are highest for the 0° crack which indicates that cracks at 0° angle to load have the low-
est chance of propagation of crack compared to others. The SIF values decrease with the
increase in crack orientation, and minimum for 90° crack which has highest chance of propa-
gation of crack. It is observed in Figure 7 that the values of SIF in sliding mode vary from
a positive number to a negative number between the crack front for all the angles except for
90° angle where in it reverses its behavior. The value of K3 plotted in the graph is shown in
Figure 8. For the cracks of all the orientations, the stress intensity factor values are observed
385

to be lower at the center of the crack front when compared to the extremes. But this behavior
reverses for crack of 90° orientation, with the values being higher at the center of the crack
front when compared to the ends. Also, the values become negative with the increase in the
crack length from extremes to center of crack front.
4 PARAMETRIC STUDIES
The parametric studies of this project provide an insight into the stress intensity factor behav-
ior at the crack front for various thermal loads and static load condition. Semi-elliptical crack
length, crack depth and crack orientations are the parameters considered for the analysis. In
the current work, how the stress intensity factor varies in three modes for cracks at different
crack orientations with different a/c ratios and different loading conditions are studied.
Thermal-Static loads: For thermal-static load case, Titanium alloy of yield strength 990 MPa,
fracture toughness 148 MPa (m)1/2 and coefficient of thermal expansion 8.1 × 10-6/°C with tem-
perature T3 = 601°C are considered.
Mode-1: The behavior of mode-1 (opening mode) stress intensity factor of thermal
and static load conditions are very similar for cracks of all orientations. From the graphs
(Figures 12–17), the values of K1 for cracks of different a/c ratios proportionally decrease
with the increase in crack orientations. There is a notable variation in the proportional
decrease of K1 value with the increase in crack orientation i.e. less reduction at one end of
crack front in comparison with the other end where the variation is large. This can be attri
buted to the curvature and twist along the span of the blade, leading to K1 values which result
in crack closure due to compression. Maximum value of K1 for thermal load case is reduced
nearly 5 times in comparison with static load case. The value of K1 becomes maximum for
larger value a/c ratio in static load case where as in thermal load case the value of K1 increase
from crack angle 0° to 30° and keeps on decreasing for higher crack angle. This shows that
cracks with a larger length when compared to the depth have a higher tendency of propaga-
tion near the crack tip and cracks which are comparatively smaller in length have a higher
tendency of propagation near the Centre of crack front, leading to component failure.
The Figures 9 to 13 show that the values of K1 changes from along the crack length at 0°
crack angle to diagonal way continuously with the increase in crack angle. The extreme values
of K1 are positive in static load case where as in thermal load case, the value of K1 changes
Figure 9. K1 values for 0° crack angle. Figure 10. K1 values for 30° crack angle.
Figure 11. K1 values for 60° crack angle. Figure 12. K1 values for 75° crack angle.
386

Figure 13. K1 values for 90° crack angle.
Figure 14. K1 values for 0° crack. Figure 15. K1 values for 30° crack.
Figure 16. K1 values for 60° crack. Figure 17. K1 values for 75° crack.
from a positive value to a negative value between the two crack fronts except for a crack angle
0°. Two important parameters; semi-elliptical crack length and depth are considered for the
analysis, and variation in the stress intensity factors in three modes for cracks at different
orientations and different a/c ratios are studied in the present work.
The Figures 9 to 13 show that cracks with a larger length when compared to the depth have
a higher tendency of propagation, leading to component failure.
The behavior of K1 values for cracks of orientations 60° and 75° is observed to be differ-
ent from the rest. The K1 values, from Figures 9–13, are negative for these two cracks. This
can be attributed to the curvature and twist along the span of the blade, leading to K1 values
which result in crack closure due to compression.
As seen from the Table 2 for both static and thermal load case, extreme values of K1 for all
the orientation of the crack is located near the proximity of crack tip except for crack angle
0° where the maximum value of K1 is located at the centre of the crack length. For static load
case, (K1) max = 30 Mpa √m at a/c ratio 0.80 and the crack orientation 0°. Similarly (K1)
min = 3.6 Mpa √m at a/c ratio 0.40 and the crack orientation 90°. Extreme values of K1 are
decreasing proportionally with the increase of crack orientation.
Thermal load case, (K1) max = 6.6 Mpa √m at a/c ratio 0.80 and the crack orientation 30°.
Similarly (K1) min = -3.2 Mpa √m at a/c ratio 0.40 and the crack orientation 30°. Maximum
value of K1 initially increases from crack angle 0° to 30° and decreases with the further
387

Table 2. Extreme values of K1 for mode-1 case.
Stress intensity factor for mode-1 (K1) in Mpa √m
Crack orientation in deg 0 15 30 45 60 75 90
(K1) max in Mpa √m 30 28 23 14.8 9.2 6.8 7.2

Crack length in mm 0 0 0 0.10 0.15 0.15 0.10
a/c Ratio 0.80 0.80 0.80 0.80 0.80 0.80 0.80
Static 3.6
(K1) min in MPa √m 18 16 11.5 7.6 5.0 3.7
Crack length in mm 0 2 2 2 2 2 2
a/c Ratio 0.40 0.40 0.40 0.40 0.40 0.40 0.40
(K1) max in Mpa √m 1.6 4.4 6.6 6.4 5.8 4.8 3.8
Crack length in mm 1.3 0 0 0.10 0.10 0.10 0.10
Thermal a/c Ratio 0.40 0.80 0.80 0.80 0.80 0.80 0.80
(T3 = 601°C) (K1) min in MPa √m 0.7 -2.0 -3.2 -2.2 -0.8 0.15 0.2
Crack length in mm 0 2 2 2 2 2 2
a/c Ratio 0.40 0.40 0.40 0.40 0.40 0.40 0.40
Table 3. Extreme values of K1 for mode-1 case.
Stress intensity factor for mode-1 (K1) in Mpa √m
Thermal load →
Parameters ↓ T1 = 500°C T1 = 550°C T1 = 601°C
(K1) max in MPa √m 9.6 8.2 6.6

Maximum Crack length in mm 0 0 0
value a/c Ratio 0.80 0.80 0.8
Crack orientation in deg 30 30 30
(K1) min in MPa √m -0.6 -2.0 -3.2
Minimum Crack length in mm 2 2 2
value a/c Ratio 0.40 0.40 0.40
Crack orientation in deg 30 30 30
Figure 18. K1 values for 90° crack.
increase of crack angle and minimum value of K1 initially decreases from crack angle 0° to
30° and increases with the further increase of crack angle.
As seen from the Table 3 for both static and thermal load case, extreme values of K1 for all
the orientation of the crack is located near the proximity of crack tip except for crack angle
0° where the maximum value of K1 is located at the centre of the crack length. For static
load case, (K1) max = 30 Mpa √m at a/c ratio 0.80 and the crack orientation 0°. Similarly
388

(K1) min = 3.6 Mpa √m at a/c ratio 0.40 and the crack orientation 90°. Extreme values of K1
are decreasing proportionally with the increase of crack orientation.
Maximum value of K1 initially increases from crack angle 0° to 30° and decreases with the
further increase of crack angle and minimum value of K1 initially decreases from crack angle
0° to 30° and increases with the further increase of crack angle as shown in the Figures 14–18.
5 CONCLUSIONS
The following conclusions are drawn from the above work.
Thermal-static loads
Maximum value of K1 for thermal load case is reduced nearly 5 times in comparison with
static load case. From the FEA results, for static load case, it can be concluded that fracture
by mode-1 is likely to occur because the values of K1 and it is found at 0° crack orientation.
The maximum values K1 and K2 for thermal load case are found at crack angle 30° and 0°
respectively.
For the temperature from T1 = 500°C to T2 = 550°C, maximum value of K1 decrease by
14.6% and further increase of temperature from T2 = 550°C to T3 = 601°C, maximum value
of K1 decrease by 19.5%. From FEA results thermal load case, it is observed that with the
increase of temperature, the value of K1 and K3 decreases in proportion along the crack
length whereas the value of K2 increases with the increase of temperature.
ACKNOWLEDGMENT
We would like to express my sincere gratitude to TEQIP III and Management B.M.S. College
of Engineering, Bengaluru, for extending financial support in publishing and presenting the
paper.
REFERENCES
ASM Aerospace Specification Metals Inc., 13 May 2016. Available: http://asm.matweb.com/search/

SpecificMaterial.asp?bassnum=MTP641.
Barlow, K.W. & Chandra, R. 2005. Fatigue crack propagation simulation in an aircraft: International
Journal of Fatigue 27, pp. 1661–1668.
Kyo-Soo Song, Seon-Gab Kim, Daehan Jung & Young-Ha Hwang, 2007. Analysis of the fracture of a
turbine blade on a turbojet engine: Engineering Failure Analysis 14, pp. 877–883.
Lucjan Witek. 2011. Stress intensity factor calculations for the compressor blade with half-elliptical
surface crack using Raju-Newman solution: Fatigue of Aircraft Structures, Vol. 1, pp. 154–165.
Nalla, et al., R.K. 2002. Mixed-mode, high cycle fatigue-crack growth thresholds in Ti-6Al-4V: Role of
small cracks: International Journal of Fatigue.
Newman, J.C. & Raju, I.S. 1981. An empirical stress-intensity factor equation for the surface crack:
Engineering Fracture Mechanics Vol. 15, No. 1–2, pp. 185–192.
Pierre Ladevèze, T. 2016. Advanced Modeling and Simulation in Engineering Sciences: Springer Open,
3.30.
Wilson Frost, et al., 2003. Fourth-stage turbine bucket airfoil: U.S. Patent 6503059 B1.
389

Numerical models and computational methods

Computational Fluid Dynamics (CFD) simulation

of gas-liquid-solid three-phase fluidized bed
G.P. Deepak & K. Shaji

Government Engineering College, Kozhikode, Kerala, India
ABSTRACT: Three-phase fluidized beds have much significance since they offer
excellent heat and mass transfer rates. Hence they are utilized in major industries such as
biotechnology, pharmaceuticals, food, chemicals, and environmental and refining plants.
Computational Fluid Dynamics (CFD) is an economical method by which to study the
hydrodynamic properties of three-phase fluidized beds, because experimental and theoretical
methods have their own limitations. A study of the hydrodynamics of a gas-liquid-solid
(three-phase) fluidized bed has been made using ANSYS Fluent simulation software. The
simulation has been carried out on a cylindrical column 1.8 m tall and 0.1 m diameter. Glass
particles of diameter 2.18 mm and 3.05 mm were used for initial bed heights of 0.267 m and
0.367 m. The hydrodynamic properties, such as bed expansion, holdup of all three phases,
and pressure drop across the column, were studied by varying inlet water velocity and inlet
air velocity. Finally, comparison was made between the results obtained from the simulation
and the experimental results. The CFD simulation result shows excellent agreement with the
experimental results.
Keywords: Computational Fluid Dynamics, three-phase fluidized beds, simulation
1 INTRODUCTION
Fluidization is achieved by a continuous supply of fluid on a bed of solid particles. The

upward drag force exerted on the solid particles by the fluid becomes exactly the same as the
gravitational force at a particular value of fluid velocity and thus keeps the particles in a sus-
pended state inside the fluid. After reaching this condition the bed has achieved a fluidized
state and exhibits fluid-like behavior. The velocity of the fluid where fluidization is achieved
is considered as the superficial fluid velocity.
Gas-liquid-solid fluidization uses three different phases where a bed of solid particles are
held in the medium of two different fluids, as shown in Figure 1. This leads to excellent con-
tact between different phases in the system.
Saha et al. (2016) presented a 2-dimensional Computational Fluid Dynamics (CFD) study
of a three-phase fluidized bed. They studied holdups of liquid and gas in the column. They
found that the gas holdup varies inversely with liquid velocity. By contrast, holdup of gas was
found to vary directly with air velocity. Witt et al. (1998) implemented a multiphase Eulerian–
Eulerian approach in estimating the transient performance of a fluidized bed. A body-fitted
grid system was used to prevent the problem of exact geometrical representation. They studied
the isothermal flow in a 3-D bubbling fluidized bed. Li et al. (1999) proposed a 2-dimensional
model to study three-phase fluidization using a Eulerian–Lagrangian approach. The flow of
gas was represented by a volume of fraction method, the flow of solid was represented by the
Eulerian method, and the dispersed particle method was used to represent the liquid phase.
They studied bubble wake behavior in the fluidized bed. Zhang and Ahmadi (2005) proposed
393

Figure 1. Three-phase fluidization.
a 2-D model for a three-phase slurry reactor by using the Eulerian–Lagrangian approach.
They investigated the transient characteristics of flow of all the phases.
Sivalingam and Kannadasan (2009) carried out an experiment on a three-phase fluidized
bed. They made an effort to study the relationship between fluid flow rates and hydrodynamic
characteristics. The conclusion drawn from the experiment was that the gas flow rate influ-
ences the design of a fluidized bed. Sau and Biswal (2011) conducted an experimental study
and a 2-dimensional CFD study. They made an attempt to study and compare the hydrody-
namic properties of a two-phase tapered fluidized bed using the Eulerian approach. They
concluded that 3-dimensional study would provide better results than 2-dimensional study.
Mohammed et al. (2014) presented an experimental study on a three-phase fluidized bed by
considering water, kerosene and spherical plastic particles as different phases. From the experi-
ment, they found that the holdup of the dispersed phase increases with its velocity as well as
particle size. However, it decreases with continuous phase velocity. Jena et al. (2008) presented
an experimental study on a three-phase fluidized bed. They considered glass beads of differ-
ent diameters as the solid phase. Liquid was taken as continuous phase. They found that the
holdup of gas increased with gas velocity and it increased when particle size was increased.
From the brief review above of the literature, it is clear that several experimental studies
have been done on three-phase fluidized beds. It was concluded that only a few works have
been done on three-phase fluidized beds using CFD. For better comparison with experimen-
tal results, 3-dimensional is mostly preferred over 2-dimensional simulation. This is because
3-dimensional simulation gives more realistic results. Studying the three-phase fluidized bed
using CFD is found to be a promising approach, with reduced cost and effort when com-
pared to experiments.
In this work, the important hydrodynamic properties of a three-phase fluidized bed have
been studied using CFD. For this, ANSYS Fluent simulation software is used to model and
solve the problem using a Eulerian approach. The simulation results are validated by com-
parison with experimental results.
2 COMPUTATIONAL MODEL AND BOUNDARY CONDITIONS
A cylindrical column of 1.8 m height and 0.1 m diameter has been considered for the study.
Glass, water and air are taken as solid, liquid and gas phases, respectively. Water is treated as
a continuous phase. Secondary phases are glass particles and air. A uniform velocity inlet and
pressure outlet boundary condition with mixture gauge pressure 0 Pa has been used. At the
wall, no slip condition has been used for water. X = 0, Y = 0 specified shear was used for air
and glass. The parameters used in the simulation are tabulated in Table 1.
394

Table 1. Important parameters.
Parameter Value
Diameter of the column 0.1 m

Height of the column 1.8 m
Density of glass particles 2470 kg/m3
Density of air 1.225 kg/m3
Density of water 998.2 kg/m3
Viscosity of air 1.789 × 10-5 kg/ms
Viscosity of water 0.001003 kg/ms
Mean particle size 2.18 mm, 3.05 mm
Initial bed height 0.267 m, 0.367 m
Inlet air velocity 0.02–0.08 m/s
Inlet water velocity 0.08–0.16 m/s
Initial solid holdup 0.6
2.1 Interphase momentum exchange

Interphase momentum exchange terms consist of different forces. Added mass force is negli-
gible when compared to drag force. In most cases, lift force is insignificant. Hence, simulation
was carried out by assuming that drag force acts as the only interphase momentum exchange
term.
2.1.1 Drag models

Schiller–Neumann correlation has been used for liquid–air interaction and a Gidaspow
model has been used for both solid–air and solid–liquid interactions (ANSYS, 2003).
2.2 Solid pressure

The pressure arising through the collision of solid particles was calculated using the Kinetic
Theory of Granular Flow (KTGF) (ANSYS, 2003).
2.3 Turbulence models

Transport equations for the turbulent kinetic energy k and its dissipation rate ε have been
used for the study. The turbulence has been modeled using a realizable k-ε model. A high
Reynolds number is assumed in this model (ANSYS, 2003).
2.4 Equations
2.4.1 Continuity equation
∂
(ε k ρk ) + ∇(ε k ρk uk ) = 0 (1)
∂t
where
ε g + ε1 + ε s = 1 (2)
2.4.2 Momentum equation

For the liquid phase:

∂
∂t

(
)
( ρlε lul ) + ∇ ( ρlε lulul ) − ε l∇P + ∇(ε l µeff ,l ∇ul + (∇ul )T + ρlε lg + Mi,l (3)
395

For the gas phase:

∂
∂t
(
) (
)
(
ρgε g ug + ∇ ρgε g ug ug = −ε g∇P + ∇(ε g µeff ,g ∇ug + ∇ug ( ) )+ ρ ε g−M
T
g g i,g (4)
For the solid phase:

∂
( ρsε s us ) +∇ ( ρsε s us us ) = − ε s∇P +∇ (ε s µeff,s (∇us + (∇us )T ) + ρsε s g + Mi,s (5)
∂t
3 GRID INDEPENDENT STUDY
Grid independent study has a major role in determining the time required for the simulation.
As the number of elements/cells decreases, simulation time reduces. But a decreased number
of elements leads to variation from the expected results. Hence it is important to find the
number of elements which leads to accurate results with a minimum number of iterations.
Simulations were performed for different numbers of elements. The values of pressure drop
obtained are shown in Figure 2.
From Figure 2 it is clear that as the number of cells increases the value of pressure drop
also increases. With 55390 cells, the pressure drop obtained is 1380 Pa, which is close to
the experimental result with the same boundary conditions. When the number of cells was
increased further, the variation in the value of pressure drop is smaller. Hence the mesh with
55390 cells was chosen for the calculation to minimize the simulation time.
A fluidized bed which uses liquid as primary phase, and gas and solid particles as secondary
phases has been modeled and solved using ANSYS Fluent. Bed heights of 0.267 m and 0.367 m
were used to study the fluidized bed with 2.18 mm and 3.05 mm glass beads. The fluidized bed
has been studied by varying inlet superficial velocity of water and inlet air velocity.
During the simulation of the fluidized bed, variation in the profile of the bed is observed,
as shown in Figure 3. But no major change in the profile is seen after some time. This is the
proof that the bed is in a fluidized condition.
4.1 Dynamics of individual phases

Figure 4 shows the contours of phase dynamics of individual phases, that is, glass, water and
air, after reaching quasi-steady state.
Figure 2. Values of pressure drop for different numbers of cells (Vl = 0.04 m/s, Vg = 0.02 m/s).
396

Figure 4. Contours of volume fraction of
Figure 3. Contours showing the variation of glass, water and air (Vw = 0.12 m/s, Vg = 0.02 m/s,
volume fraction of glass particles with time Hs = 0.267 m, Dp = 3.05 mm).
(Hs = 0.267 m, Vw = 0.12 m/s, Vg = 0.02 m/s, Dp
= 3.05 mm).
It is clear from the contour that the bed has reached a fluidized state. The water volume
fraction is relatively smaller in the fluidized region and that of air is greater when compared
to the remaining part.
4.2 Bed expansion

The change in bed expansion with inlet air velocity at various inlet water velocities is shown
in Figure 5. Only slight variation in bed expansion is observed with inlet air velocity. Bed
expansion increases as the inlet water velocity increases. This is because the force exerted by
liquid at lower inlet water velocities is not sufficient enough to overcome the gravitational
pull on the glass particles.
The plot of change in bed expansion with inlet water velocity at different heights of bed is
shown in Figure 6. Bed expansion increases when bed height is increased.
Figure 7 is the plot of the behavior of bed expansion with inlet velocity of water for inlet
air velocity 0.02 m/s for 2.18 mm and 3.05 mm particles at bed height 0.367 m. It can be seen
that the bed expansion is lower for higher particle size.
Figure 8 is the plot of comparison of experimental and simulation results of bed expan-
sion ratio for glass particles of diameter 3.05 mm at initial bed height of 0.267 m for air veloc-
ity of 0.02 m/s. From the comparison curve, it can be seen that the values of the simulation
make a satisfactory match with experimental results.
4.3 Phase holdups

Holdup of individual phase is the mean area weighted average of volume fraction in the flu-
idized region. Average value is taken as holdup because the values differ at different locations
of the fluidized part.
Figure 9 shows the representation of variation of air holdup with water velocity at an
initial bed height of 0.267 m for various air velocities for particles of 3.05 mm. It is clear
that air holdup decreases when water velocity is increased, and it increases with increased
air velocity.
Figure 10 shows the behavior of glass holdup with inlet water velocity for glass particles
of diameter 3.05 mm by keeping 0.267 m bed height. Glass holdup decreases when water
velocity is increased by keeping inlet air velocity constant. When air velocity is increased,
glass holdup decreases slightly.
The variation of water holdup with water velocity for glass particles of diameter 3.05 mm
at 0.267 m bed height is plotted in Figure 11. Water holdup increases when the inlet water
velocity is increased. And water holdup decreases when inlet air velocity is increased.
397

Figure 6. Change in bed expansion with water
Figure 5. Variation of bed expansion with
velocity at altered bed heights (Vg = 0.02 m/s,
inlet air velocity (Hs = 0.267 m, Dp = 3.05 mm).
Dp = 3.05 mm).
Figure 8. Variation of bed expansion ratio

Figure 7. Change in bed expansion with water with water velocity (Dp = 3.05 mm, Hs = 0.267 m,
velocity for various particle sizes (Hs = 0.367 m, Vg = 0.02 m/s).
Vg = 0.02 m/s).
Figure 9. Change in air holdup with water

velocity at various air velocities (Dp = 3.05 mm, Figure 10. Variation of glass holdup
Hs = 0.267 m). with water velocity at various air velocities
(Dp = 3.05 mm, Hs = 0.267 m).
The evaluation of experimental and simulation results of air holdup is made in Figure 12
for glass particles of 3.05 mm at 0.267 m bed height for inlet air velocity of 0.02 m/s. The
match between simulation result and experimental result is excellent.
4.4 Pressure drop

The difference of pressures at inlet and outlet is known as pressure drop. The pressure drop
variation with inlet water velocity for various inlet air velocities for glass particles of diameter
3.05 mm and 0.267 m bed height is shown in Figure 13.
The pressure drop varies directly with water velocity. At lower air velocities, the opera-
tion becomes liquid-fluid because the volume fraction of air is too low and hence the pres-
398

Figure 11. Variation of water holdup with water
velocity at various air velocities (Dp = 3.05 mm,
Hs = 0.267 m). Figure 12. Variation of air holdup with
water velocity (Vg = 0.02 m/s, Dp = 3.05 mm,
Hs = 0.267 m).
Figure 13. Change in pressure drop with

water velocity for various air velocities Figure 14. Change in pressure drop with water
(Dp = 3.05 mm, Hs = 0.267 m). velocity at various bed heights (Dp = 3.05 mm,
Vg = 0.02 m/s).
Figure 16. Variation of pressure drop with

water velocity (Vg = 0.02 m/s, Dp = 3.05 mm,
Figure 15. Change in pressure drop with water Hs = 0.267 m).
velocity for various particles (Hs = 0.367 m,
Vg = 0.02 m/s).
sure drop does not increase much. Drop in pressure varies inversely with air inlet velocity.
This is because as the inlet air velocity increases gas holdup increases, and water holdup in
the column decreases. Because the density of air is much lower than that of water, as water
holdup decreases pressure drop also decreases.
Figure 14 shows the behavior of pressure drop across the column with inlet water for
3.05 mm glass particles at 0.267 m and 0.367 m bed heights. The pressure drop increases when
the initial bed height increases.
399

Figure 15 is the plot of changes in pressure drop with inlet water velocity for various
particle sizes at 0.367 m bed height. The pressure drop increases when the diameter of glass
particle is increased.
The simulation results of pressure drop are validated with experimental results for glass
particles of diameter 3.05 mm at 0.267 m bed height for 0.02 m/s air velocity. The comparison
curves are shown in Figure 16. The simulation results differ by 12% or less from the experi-
mental results.
5 CONCLUSIONS
A study of a gas-liquid-solid three-phase fluidized bed has been made using a Eulerian–Eul-
erian approach. A 3-dimensional model having 1.8 m height and 0.1 m diameter was devel-
oped. The hydrodynamic properties, such as bed expansion, air holdup, water holdup, and
pressure drop, were studied by varying inlet water velocity, inlet air velocity, particle diameter
and bed height. The major conclusions from the study are summarized as follows:
Bed expansion is directly proportional to water velocity. It is not much affected by inlet air
velocity. It increases when static bed height is increased and reduces when particle size is
increased.
Gas holdup is directly proportional to inlet air velocity. It decreases when inlet water veloc-
ity is increased.
Glass holdup decreases when inlet water velocity is increased and only slight variation is
observed in glass holdup when inlet air velocity is increased.
Water holdup varies directly with inlet water velocity and it decreases with an increase in
air velocity.
Pressure drop varies directly with inlet water velocity. It varies inversely with inlet air
velocity. Pressure drop increases when initial bed height and particle diameters were
increased.
The simulation results show excellent agreement with experimental results.
REFERENCES
ANSYS. (2003). Fluent 6.1 User’s Guide (pp. 1–5). Canonsburg, PA: ANSYS.
Blazek, J. (2001). Computational fluid dynamics: Principles and applications (1st ed.). Oxford, UK:
Elsevier.
Jena, H.M., Sahoo, B.K., Roy, G.K. & Meikap, B.C. (2008). Characterization of hydrodynamic proper-
ties of a gas-liquid-solid three-phase fluidized bed with regular shape spherical glass bead particles.
Chemical Engineering Journal, 145, 50–56.
Li, Y., Zhang, J. & Fan, L.S. (1999). Numerical simulation of gas-liquid-solid fluidization system using
a combined CFD-VOF-DPM method: Bubble wake behaviour. Chemical Engineering Science, 54,
5101–5107.
Mohammed, T.J., Sulaymon, A.H. & Abdul-Rahmun, A.A. (2014). Hydrodynamic characteristic
of three phase (liquid-liquid-solid) fluidized beds. Journal of Chemical Engineering and Process
Technology, 5, 188.
Saha, S.N. Dewangan, G.P. & Gadhewal, R. (2016). Gas-liquid-solid fluidized bed simulation.
International Journal of Advanced Research in Chemical Science, 3, 1–8.
Sau, D.C. & Biswal, K.C. (2011). Computational fluid dynamics and experimental study of the
hydrodynamics of a gas–solid tapered fluidized bed. Applied Mathematical Modelling, 35, 2265–2278.
Sivalingam, A. & Kannadasan, T. (2009). Effect of fluid flow rates on hydrodynamic characteristics
of co-current three phase fluidized beds with spherical glass bead particles. International Journal of
ChemTech Research, 1, 851–855.
Witt, P.J., Perry, J.H. & Schwarz, M.P. (1998). A numerical model for predicting bubble formation in a
3D fluidized bed. Applied Mathematical Modelling, 22, 1071–1080.
Zhang, X. & Ahmadi, G. (2005). Eulerian–Lagrangian simulations of liquid-gas-solid flows in three-
phase slurry reactors. Chemical Engineering Science, 60, 5089–5104.
400

Fault diagnosis of self-aligning troughing rollers in a belt

conveyor system using an artificial neural network and Naive
Bayes algorithm
S. Ravikumar
GKM College of Engineering and Technology, Tamilnadu, India
S. Kanagasabapathy
National Engineering College, Kovilpatti, Tamilnadu, India
V. Muralidharan
B.S. Abdur Rahman University, Chennai, Tamilnadu, India
R.S. Srijith & M. Bimalkumar

GKM College of Engineering and Technology, Tamilnadu, India
ABSTRACT: The belt conveyor system is used for conveying large volumes of materials
from one location to another. The Self-Aligning Troughing Roller (SATR) is one of the critical
components in the belt conveyor; it is quite influential in riding the belt conveyor in fault free
conditions. SATR has to operate under heavy axial and shear forces, which lead to frequent
failures. Hence continuous monitoring and fault diagnosis of SATR becomes essential. The
self-aligning troughing idler arrangement has a long roll to support the belt and handle maxi-
mum loads per cross-section. The self-aligning troughing roller has machine elements, includ-
ing the ball bearing, a central shaft and the external shell. In the belt conveyor system certain
faults, such as Bearing Flaws (BF), Central Shaft Faults (CSF), combined Bearing Flaws and
Central Shaft Faults (BF & CSF) occur frequently. A prototype investigational model has been
made with the above mentioned faults and the vibration signals were attained from the set-up.
The vibration data acquired was fed as algorithm input into Artificial Neural Networks (ANN)
and Naive Bayes (NB) algorithms, which are used for classification of acquired signals. In the
present effort, the artificial neural networks and Naive Bayes algorithms were found to achieve
82.1 and 90% classification accuracy, which acknowledges that the Naive Bayes algorithm has
a better gain over artificial neural networks in the field of fault diagnosis applications.
1 INTRODUCTION
The self-aligning troughing roller (SATR) is an essential element of the belt conveyor system.
It may fail due to multidimensional forces, derisory lubrication, culpable sealing, uneven
loading and improper training of belt. The critical elements that fail periodically in the self-
aligning troughing roller are the groove ball bearing and the central shaft.
The malfunction of these parts directly affects the efficiency of the SATR, which can
hinder the proper functioning of the belt conveyor system. In these circumstances, to avoid
overwhelming damage of the belt conveyor, a failure prediction system is a major require-
ment. The various conditions for this research are SATR running in Fault Free Condi-
tion (FFC), Bearing Fault Condition (BFC), Central Shaft Fault (CSF) and Bearing Fault
and Central Shaft Fault (BFC & CSF). The malfunction of these components affects the
functioning of SATR which in turn leads to under-performance of the belt conveyor system.
401

The conventional and Fast Fourier Transform (FFT) methods work well when signals are
static, but in continuous varying conditions they are ineffective. Here the components of SATR
generate vibration signals with significant variation. These signals are acquired and the selection
features are extracted from it. The selection of these features is based on their impact in fault
prediction, which is the consequent stage of SATR condition monitoring. Apart from this, a
good quality fault diagnosis algorithm tool has to be utilized for classification. At present there
are a fair number of classification algorithms, each having their own pros and cons.
In the present case artificial neural networks (ANN) and Naive Bayes (NB) algorithms are
selected for analysis. The classification correctness differs among algorithms, likewise ANN and
NB were found to achieve different performance levels. Thus, it is necessary to select a relevant
algorithm that can be used to assess the condition monitoring and fault diagnosing of SATR.
2 RELATED WORK
Murru (2016) presented an original algorithm for initialization of weights in back propa-
gation neural net with application to character recognition. The initialization method was
mainly based on a customization of the Kalman filter, translating it into Bayesian statistics
terms. A metrological approach was used in this context considering weights as measure-
ments, modeled by mutually dependent normal random variables. The algorithm perform-
ance was demonstrated by reporting and discussing results of simulation trials. Results were
compared with random weights initialization and other methods. The proposed method
showed an improved convergence rate for the back propagation training algorithm.
Wong et al. (2016) proposed a Probabilistic Committee Machine (PCM), which combines
feature extraction, a parameter optimization algorithm and multiple Sparse Bayesian Extreme
Learning Machines (SBELM) to form an intelligent diagnostic framework. Results showed
that the proposed framework was superior to the existing single probabilistic classifier. Zhang
et al. (2016) developed a Bayesian statistical approach developed for modal identification using
the free vibration response of structures. The results indicated that a frequency-domain Baye-
sian framework was created for modal identification of Most Probable Values (MPVs) and
modal parameters Mori & Mahalec (2016) introduced a decision tree structured conditional
probability representation that can efficiently handle a large domain of discrete and continuous
variables. Experimental results indicated that our method was able to handle the large domain
discrete variables without increasing computational cost exponentially.
Hu et al. (2016) developed the framework of Non-Negative Sparse Bayesian Learning
(NNSBL). The algorithm obviated pre-setting any hyper parameter, where the Expectation
Maximization (EM) algorithm was exploited for solving this NNSBL problem. Without a
prior knowledge of the source number, the proposed method yielded performances in the
underdetermined condition illustrated by numerical simulations.
Kiaee et al. (2016) utilized the concept of random effects in the Extreme Learning Machine
(ELM) framework to model inter-cluster heterogeneity, provided the inherent correlation
among the samples of a particular cluster is taken into account, as well. The proposed ran-
dom effect model includes additional variance components to accommodate correlated data.
Inference techniques based on the Bayesian evidence procedure were derived for the estima-
tion of model weights, random effect and residual variance parameters as well as for hyper
parameters. The proposed model is applied to both synthesis and real-world clustered data-
sets. Experimental results showed that our proposed method can achieve better performance
in terms of accuracy and model size, compared with the previous ELM-based models.
Wang et al. (2016) used the Gaussian kernel function with smoothing parameter to
estimate the density of attributes. A Bayesian network classifier with continuous attributes
was established by the dependency extension of Naive Bayes classifiers. The information
provided to a class for each attributes as a basis for the dependency extension of Naive Bayes
classifiers is analyzed. Experimental studies on UCI datasets showed that Bayesian network
classifiers using Gaussian kernel function provided good classification accuracy compared
to the approaches when dealing with continuous attributes. Magnant, et al. (2016) proposed
402

Bayesian non-parametric models. It showed the possible functional forms of the state noise
covariance matrices, which eliminate the number of time switching hyper parameters in many
applications. Results showed that the online estimation of the state noise precision matrix
used by DP methods allowed improvisation of tracking accuracy.
Muralidharan & Sugumaran (2016) have introduced a more logical algorithm for fault diag-
nosis of the centrifugal pump. The extracted features were classified and analyzed by SVM and
Extreme Learning Machine (ELM). Finally it is concluded that ELM has higher accuracy than
SVM. Muralidharan et al. (2015) elucidated, that the centrifugal pump fault can be identified
using the pattern detection method. The input signal is extracted through Stationary Wavelet
Transform (SWT) and classification is made through Bayes net. Finally it is concluded that the
Bayes net is better for fault diagnosis of the centrifugal type monoblock pump.
3 EXPERIMENTAL SET-UP
SATR fault diagnosis involves several steps regarding the conveyor set-up: (i) design and
fabrication with multiple fault conditions, (ii) acquisition of signals, (iii) feature extraction
and (iv) feature classification. The procedure of the process can be clearly understood from
Figure 1. Initially, the belt conveyor model is allowed to run with parts working in a fault
free condition and the signals were acquired. One set of shaft and bearing are prefabricated
with faults. As given in Table 1, the outer ring thickness of 4.5 mm and 4.51 mm respectively
were ground for 4 mm and 3.90 mm respectively for developing faults in the groove bearings.
The bearing was attached to a self-aligning troughing roller set-up in the conveyor system.
Similarly the central shaft was ground to create shaft fault as shown in Table 2.
Figure 1. Flowchart for fault diagnosis of belt conveyor.
403

Table 1. Groove bearing diameter readings.
Bearing outer ring thickness mm
Sl. No. Before grinding After grinding
1 4.50 (LHS side) 4.0

2 4.51 (RHS side) 3.90
Table 2. Central shaft diameter readings.
Diameter of shaft mm
Before After
Sl. No. grinding grinding Side
1 10.01 9.88 Left Side

2 10.00 9.90
3 10.00 9.98 Right Side
4 9.99 9.99
Figure 2. Experimental set-up.
The roller bearing (Model No KG6200Z) and central shaft were prefabricated with a fault
to acquire the vibration readings as given in Tables 1 and 2. The different fault conditions, like
SATR having Bearing Fault (BF), central shaft fault (CSF) and combined central shaft and
bearing fault (CSF & BF) were subsequently set up one by one and the corresponding vibra-
tion signals were acquired. Figure 2 shows a schematic arrangement about the SATR vibra-
tion analysis experimental set-up. A piezoelectric accelerometer sensor (Model No 3055B1)
was mounted over the vibration zone to absorb the vibrations generated due to faults.
A signal conditioning unit was connected to the accelerometer sensor which in turn was
connected to an Analog-to-Digital Converter (ADC). The digital vibration signals acquired
from the ADC were fed to the computer for further processing through relevant tools. The
LabVIEW software was used to record the vibration signals in the digital form and store them
in the computer hard disk memory. It is further processed different features were extracted by
add-on in Microsoft Excel.
4 FEATURE EXTRACTION
Different statistical parameters are used to show the fault conditions in the fabricated model.
The various statistical parameters include mean, median, mode, standard error, standard
deviation, kurtosis, skewness, minimum value, maximum value, sample variance and range.
404

Table 3. Individual feature classification accuracy.
Individual feature classification accuracy (%)
Standard deviation 71
Skewness 63.45
Sample variance 50
Standard error 44.66
Range 38.01
Figure 3. Sample plots of selected features with different conditions.
4.1 Selection of important features

In this stage, eight common features were chosen from the acquired vibration signals in order
to see the difference between fault and fault free conditions. The classification accuracy of
each feature was noted down. The feature with high accuracy of 71% was found be with
standard deviation. The next highest value of 63.45% was found to be with skewness. Hence,
when the second feature was combined with the first one, then the classification accuracy
followed an increasing trend. Similarly, the next highest value was found to be with sample
variance and hence it was combined with the other two higher accuracy features.
The trend was found to be increasing. The procedure was repeated for all the features sepa-
rately and the features with more competence to classify the faults were identified.
The selected features are standard deviation, skewness, sample variance, and standard
error, which have high classification accuracy over the remaining statistical features as can
be seen in Figure 3.
5 CLASSIFIER
5.1 Artificial neural network

Artificial neural network (ANN) is a method that has been developed with the concept of
genetic neurons of the nervous systems. It is capable of self-interpreting the signals that are
provided as inputs.
It has hidden layers between the input and output layers. The input signals are learned and
interpreted by means of activation functions, which are associated with every hidden layer.
Usually, the processing is performed by multilayer perceptrons.
405

5.2 Multi-Layer Perceptron (MLP)
An MLP can be viewed as a logistic regression can learn nonlinear transformation. This
transformation projects the input data into a space where it becomes linearly separable. This
intermediate layer is referred to as a hidden layer. A single hidden layer is sufficient to make
MLPs a universal approximator. However, we will see later on that there are substantial ben-
efits to using many such hidden layers, i.e. the very premise of deep learning. Naive Bayes.
5.3 Naive Bayes

Naive Bayes method is a powerful algorithm for the classification of sample by a set of super-
vised learning algorithms. It is applied based on Bayes’ theorem with the “naive” assumption
of independence between every pair of features. Given a class variable y and a dependent
feature vector x 1 through x n, Bayes’ theorem states the following relationship:
 P ( y / x1,… xn / y ) 
P ( y / x1,… xn ) = P ( y )P  (1)
 P ( x1,… xn ) 

It can be used for a Maximum A Posteriori (MAP) estimation to estimate P(y) and P( x i \y );
the former is then the relative frequency of class y in the training set. The Naive Bayes clas-
sifiers differ mainly by the assumptions they make regarding the distribution of P( x i \y )
in spite of their apparently over-simplified assumptions. Naive Bayes classifiers have worked
quite well in many real-world situations, particularly for document classification and spam
filtering. They require a small amount of training data to estimate the necessary parameters.
Naive Bayes learners and classifiers can be extremely fast compared with more sophisticated
methods. The decoupling of the class conditional feature distributions means that each dis-
tribution can be independently estimated as a one dimensional distribution. This in turn
helps to alleviate problems stemming from the curse of dimensionality.
The vibration signals were acquired for the various faults, like fault free condition, central
shaft fault, bearing fault condition and combined central shaft and bearing faults. A total of
250 data points were taken for each fault condition. Furthermore, the sample was split into
two equal parts with the first phase for training followed by testing. In each phase 125 signals
were taken for classification.
In section 4 and 4.1, feature extraction, feature selection has been discussed. Of the exist-
ing statistical eleven features were suggested for classification. However, the best ones were
selected for the following reasons:
To avoid unnecessary computation and poor results (dimensionality reduction).
To save time and build the robust model.
For the classification and validation Naive Bayes and ANN algorithm have been utilized
and the results were discussed. The effectiveness of the sample quality is understood by the
True Positive (TP) rate and False Positive (FP) rate in Table 4. A quality classification has a
value approaching ‘0’ for the false positive (FP) and for true positive the value has to be near
to 1. In Table 4 the TP value is very significant as it approaches 1 and the FP value is near
0, which highlights the quality of this classification. In addition, the classified data may be
exhibited in the form of a confusion matrix as indicated in Table 5.
The important features that are highly participative in deciding the various faults of the
SATR are standard deviation, skewness, standard variance, and standard error is vital in
deciding the faults. The standard deviation determines how much variability is sorted in a
coefficient estimate. A coefficient is significant if it is non-zero. Standard deviation is used
to measure the number of faults around and non-faulty conditions. The higher the standard
406

Figure 4. Comparison of NB and ANN for different number of features.
Table 4. Individual feature classification accuracy.
TP rate FP rate Precision Recall F measure ROC area Class
0.945 0.062 0.738 0.745 0.853 0.861 FFC

0.854 0.073 0.053 0.654 0.672 0.836 CSF
1 0 1 1 1 1 BFC
0.841 0.036 0.891 0.843 0.824 0.885 BFC & CSF
Weight Avg. 0.882 0.047 0.784 0.782 0.782 –
Table 5. Confusion matrix.
FFC CSF BFC BFC & CSF
FFC 212 38 0 0
CSF 33 193 0 22
BFC 0 2 248 0
BFC & CSF 0 14 0 236
deviation value the larger the gap between faulty conditions and good condition. Skewness
measures the symmetry in the samples.
When the skewness value reaches zero, it indicates an error free condition. The critical fea-
tures selected for classification has shown a clear margin from each other, which substantiate
the selection of these features (Figure 3). The results of the 1,000 samples were reviewed with
the help of a confusion matrix. The understanding of the confusion matrix is essential before
exposing the same which is presented in Table 5.
The fault free condition (FFC) is represented in the first row, followed by central shaft
fault (CSF) subsequently the bearing fault condition (BFC) and the combined fault condi-
tion (CSF & BFC) was shown in the third and fourth rows in the table.
It was obvious from the confusion matrix (Table 5), that 250 samples were taken for the
different conditions of the self-aligning training roller. Those diagonal elements of the con-
fusion matrix speak about the effectively ordered information and the incorrectly classified
data points are positioned as non-diagonal elements. This is how the classification accuracies
are predicted from the confusion matrix.
In this case, 212 fault free condition (FFC) data has been effectively ordered and the
remaining 36 data demonstrate as central shaft fault (CSF). Similarly, 193 data points of cen-
tral shaft fault (CSF) have been effectively ordered and 38 ineffectively ordered as fault free
conditions. In this manner, the confusion matrix is inferred and the classification precision
obtained was 90%. These results are for a particular dataset hence, the classification accuracy
of 90% may guarantee a similar performance for all similar feature data. Furthermore it is
407

possible to expect a classification accuracy close to 90% by Naive Bayes when compared to
the ANN classifier.
7 CONCLUSION
From the current analysis, it is evident that current research in a coal handling belt convey-
ors SATR has extensive scope for further application and examination. The prototype set-
up has been created with a high sensitive accelerometer of 10 Mv/g and a frequency range
of 0–2000 Hz (3dB) suitable for vibration monitoring. The accelerometer was hermetically
mounted at the self-aligning troughing roller’s stinger support (shown in Figure 2), which is
an ideal accelerometer mounting technique for vibration extraction. The vibration signals
were acquired using a data acquisition system. The acquired information has been preproc-
essed to extract those measurable features. The superlative features were distinguished utiliz-
ing the Naive Bayes algorithm and the faults were classified with them.
Since the information has been acquired in a particular working state, the end result may
be comprehensive for similar cases. Aiming at the shortcomings of the conventional failure
analysis of self-aligning troughing roller, the methodology adopted would definitely serve as
a guideline for the future research work on this area. However, the classification accuracy
of 90% is significant in this application. It can be concluded that the statistical features and
Naive Bayes algorithms are the best options for fault diagnosis of self-aligning troughing
roller in a bulk material handling belt conveyors.
REFERENCES
Hu, B., Sun, J., Wang, J., Dai, C. & Chang, Localization for sparse array using nonnegative sparse Baye-
sian learning. Signal Processing, 127, 37–43.
Kiaee, F., Sheikhzadeh, H. & Eftekhari Mahabadi, S. (2016). Sparse Bayesian mixed-effects extreme
learning machine, an approach for unobserved clustered heterogeneity. Neurocomputing 175, 411–420.
Magnant, C., Giremus, A., Grivel, E., Ratton, L. & Joseph, B. (2016). Bayesian non-parametric meth-
ods for dynamic state-noise covariance matrix estimation: Application to target tracking. Signal
Processing, 127, 135–150.
Mori, V. Mahalec, (2016). Inference in hybrid Bayesian networks with large discrete and continuous
domains. Expert Systems with Applications, 49, 1–19.
Muralidharan V., Sugumaran, V. & Sakthivel, N. (2014). Fault diagnosis of monoblock centrifugal
pump using stationary wavelet features and Bayes algorithm. Asian Journal of Science and Applied
Technology, 3, 1–4.
Muralidharan, V. & Sugumaran, V. (2017). A comparative study between Support Vector Machine
(SVM) and Extreme Learning Machine (ELM) for fault detection in pumps. Indian Journal of Sci-
ence and Technology, 9, 1–4.
Murru, N. & Rossini, R. (2016). A Bayesian approach for initialization of weights in backpropagation
neural net with application to character recognition. Neurocomputing, 193, 92–105.
Wang, R., Gao, L.-M. & Wang, (2016). Bayesian network classifiers based on Gaussian kernel density.
Expert Systems with Applications, 51, 207–217.
Wong, P.-K., Zhong, J., Yang, Z.-X. & Vong, C.-M. (2016). Sparse Bayesian extreme learning committee
machine for engine simultaneous fault diagnosis. Neurocomputing, 174, 331–343.
Zhang, J. Chen, Z. Cheng, P. & Huang, X. (2016). Multiple-measurement vector based implementation
for single-measurement vector sparse Bayesian learning with reduced complexity, Signal Processing,
118, 153–158.
408

Other topics related to mechanical engineering

Structural analysis of midship section using finite element method
Sirosh Prakash
Department of Naval Architecture and Shipbuilding Engineering, Sree Narayana Gurukulam College
of Engineering, Kolenchery, Ernakulam, India
K.K. Smitha
Department of Civil Engineering, Sree Narayana Gurukulam College of Engineering, Kolenchery,
Ernakulam, India
ABSTRACT: The design used in the present work is from the conceptual design of a
container ship and includes preliminary analysis of the structural design of the midship
section. The main objective of the work is to study the structural response of the midship
section to static loading. The present work is carried out in ANSYS, which is well-known
finite element modeling software. The first step is to produce the required model in ANSYS,
which we can either model in ANSYS or can import from any CAD software. For this
analysis, the model has been developed in ANSYS Parametric Design Language (APDL).
As it is a preliminary analysis, there are many assumptions, such as reducing the number
of girders and longitudinal stiffeners. The different static forces taken into account are the
hydrostatic pressure acting on the ship’s hull, the self-weight of structural members, and the
loads from containers. There are two approaches for load application: the load combination
method and the resultant force method. In the first method, we apply the forces acting on
each structural member. In the second method, we find the resultant forces and these are
applied on some respective structural members.
Keywords: structural analysis, finite element method, midship
1 INTRODUCTION
Ships when exposed to sea will undergo different kind of forces, including deformation, so it
is necessary do a preliminary analysis to find the response of ship structures. Generally, there
are two types of loads acting on a ship, dynamic loads and static loads. Ship structural design
is a challenging task during the shipbuilding process. The structural design should fulfill two
main objectives. One is to design the ship structure to withstand the loads acting on it; the
other is to design the structural members economically.
The next step in structural design is the evaluation of loads, nature of loads, and so on. The
initial structural dimensions are fixed according to stress analysis of beams, plates and
the shell under hydrostatic pressure, bending and concentrated loads. The loads that strongly
affect the deformation of hull girders are hydrostatic pressure and cargo loads (Eyres, 1988;
Taggart, 1980; Souadji, 2012).
2 MODEL DEVELOPMENT IN ANSYS SOFTWARE
The modeling procedure using ANSYS software can be divided into three parts, that is, the
modeling of the side shell, the modeling of side girders, and the modeling of the bulkhead.
The following steps illustrate the process of model development:
411

1. First, it is necessary to specify the type of analysis, whether it is thermal, structural, mag-
netic or electric. For our model, we have chosen structural analysis.
2. The next step is to specify the element. This software tab includes different types of ele-
ments: solid, beam, shell and link. For 3D modeling, an 8-noded brick element has been
chosen.
3. For modeling the midship structure, the side shell must first be created, and the key point
must be defined. The key point is positioned according to the coordinate values of the side
shell model.
4. After modeling the side shell, the bottom structure, including center girders and side
girders, has to be modeled. The inputs for modeling the girders are the thickness of the
plate and the height of the girders. The four key points of the girders are provided. These
four key points are joined using the Line command in modeling. The plane obtained is
extruded to the required dimensions.
5. The modeling of the bulkhead starts by modeling a plate, which is obtained by defining
key points that are joined by lines to create a default area. This area is extruded about an
axis to model the plate. Once the plate for the bulkhead is created, modeling of vertical
frames and horizontal stiffeners are carried out (Souadji, 2012).
3 MESHING IN ANSYS
In meshing, the surface and volumes are divided into a number of elements using nodes.
The accuracy of the meshed model varies according to the element selected (e.g. 10-node,
20-node and 4-node quad). The accuracy also depends on the type of meshing—coarse or
fine. A coarse mesh gives poor results, because the effect of continuity is reduced. Hence,
to obtain near-perfect results it is necessary to use a fine mesh. The mesh can be modified
using the Refine option. The mesh is refined with respect to the nodes, elements, area, and
volumes. It is refined near to the edges and the joints to obtain better results. Figure 1 shows
a meshed model of a structure.
Figure 1. Meshed model in ANSYS.
412

4 APPLYING LOADS IN ANSYS
Once the meshing is completed we can apply various loads on the structures of ship. The side
shell of the ship will be subjected to hydrostatic pressure. This pressure can be calculated using
the formula ρgh, where ρ is the density of water, g is the acceleration due to gravity, and h is the
height of the water. This can be applied as a pressure load to the bottom and side shell up to the
draft level. Using ANSYS, we can apply pressure loads, forces, moments, uniformly distributed
loads (UDLs), and so on. The boundary conditions can be applied by specifying the displace-
ments UX, UY and UZ, and the rotations RX, RY and RZ. The upper side of the bulkhead
will be under the loads from decks. These loads can be applied as a uniformly distributed load.
The forces are applied to selected elements by selecting the nodes, areas, and elements.
5 APPLYING LOADS IN ANSYS
Analysis of a container ship carried out by Souadji (2012) has been taken as the basis for the
present analysis. The details of ship cross section are shown in Table 1 and Figure 2.
Structural analysis in ANSYS involves various steps, including modeling, meshing, load
application, obtaining the solution, and understanding the output. The various steps used for
the present structural analysis were as follows:
a. Assigning the type of elements used; for this work, shell type element SHELL 4 NODE
181 was used.
b. Assigning the material properties, the modulus of elasticity, Poisson’s ratio, and density.
Defining these properties helps the software to select the appropriate material.
c. Providing thickness value of the shell element. There is a separate library called Sections
to provide this thickness value.
d. In modeling, the coordinate values are provided. As the model becomes more complex,
the number of key points also increases. The key points are joined using lines, which will
give us the required area.
e. Once the modeling is completed the next process is to develop a meshed model. This is the
process in which the geometry is converted into a number of elements. Meshing is neces-
sary before the application of loads.
f. Various loads are then applied to the structure. The different static loads acting are the hydro-
static pressure, loads from containers, and the self-weight of structural members. The resultant
load is calculated and applied to the respective parts of the midship section. Load calculations:
i. Calculation of loads from containers:
Payload from one container = volume of the container × density of the mate-
rial stored = 0.454 × 80350 kg
No. of containers in a compartment = 70
Total load from containers = 2000 N/compartment
Total load including self-weight = 3000 N/compartment
ii. Hydrostatic pressure = 105581 kg/m2
Table 1. Ship particulars.
Length overall, LOA (m) 220.5

Length between perpendiculars, LBP (m) 210.2
Breadth, B (m) 32.24
Depth to main deck, D (m) 18.70
Draft scantling, T (m) 12.15
Deadweight capacity (t) 41850
Block coefficient (CB) 0.67
Speed at design draft V (kn) 22.30
Tonnage about (dt) 35881
413

Figure 2. Midship section of container ship.
g. Application of boundary conditions is an important criterion. This is done using the

Apply loads tab.
h. The next step is to solve the geometry. The solution gives the stress and displacement
components in x, y and z directions.
6 DISCUSSION OF RESULTS
Figure 3 shows the deflected profile of the structure. The maximum value of deflection
is derived as 0.84 m. Due to the complexity of the structure and the high computational
time necessary to solve the model, some of the longitudinal members are omitted from the
ANSYS model. This reduces the stiffness of the modeled structure compared to the actual
physical structure and hence a higher value of deflection is obtained in the finite element
analysis. In order to obtain a value for deflection that is closer to that of the physical model,
the full structure must be modeled mathematically in ANSYS, which shows the importance
of precise modeling of physical structures. Figure 3 also shows the stress in the x direction.
The maximum stress value is 0.301 × 109 N/m2; this value occurs at one or two points where
the stress concentration comes into the picture. This can be avoided by refining the mesh near
the stress concentration points. The stresses in x, y and z directions can be obtained from the
ANSYS software.
414

Figure 3. Stresses in X direction of the midship section.
7 CONCLUSION
Structural analysis of a midship section can be carried out using ANSYS finite element soft-
ware. The deflection and stresses in the x direction can be evaluated.
REFERENCES
Eyres, D.J. (1988). Ship construction (3rd ed., pp. 201–320). Oxford, UK: Butterworth-Heinemann.
Larsson, R. (1988). Ship structures—Basic course (MMA130). Gothenburg, Sweden: Department of
Applied Mechanics, Chalmers University of Technology. Retrieved from http://www.am.chalmers.
se/∼ragnar/ship_structures_home/lectures/L1.pdf.
Souadji, W. (2012). Structural design of a containership approximately 3100 TEU according to the con-
cept of general ship design B-178 (Master’s thesis, Western Pomeranian University of Technology,
Szczecin, Poland). Retrieved from http://m120.emship.eu/Documents/MasterThesis/2012/Wafaa%20
Souadji%20.pdf.
Taggart, R. (Ed.). (1980). Ship design and construction (pp. 130–224). New York, NY: The Society of
Naval Architects & Marine Engineers.
415

Experimental analysis of properties of a biolubricant derived

from palm kernel oil
K. Sandeep
Department of Mechanical Engineering, Karunya University, Coimbatore, Tamil Nadu, India
M. Sekar
GMR Institute of Technology, Srikakulam, Andhra Pradesh, India
ABSTRACT: Lubricants with natural origin are known for their biodegradability and hence
are called biolubricants. This study examined the tribological, physical and chemical properties
of a biolubricant derived from Palm Kernel Oil (PKO). Zinc dialkyldithiophosphate (ZDDP)
is used as an additive and comparison of the properties of this newly developed oil with pure
PKO and SAE 20W40 engine oil were conducted. Friction and wear tests were performed in
a four-ball tribo tester as per ASTM D4172 standards. Test results reveal that pure PKO has
good tribological and physical properties (with the exception of its melting point) compared
to other pure vegetable oils. Modification of PKO with ZDDP made the results even better—
values of wear scar diameter and coefficient of friction were lower than for SAE 20W40 oil.
The melting point also reduced to 9°C, which can be further reduced by chemical modification.
Keywords: Experimental analysis, biolubricant, palm kernel oil
1 INTRODUCTION
Lubricating oils are used in domestic and industrial processes to increase the life of machin-
ery. They also make the provision of energy easier and at lower cost. Growing consumption
of different lubricant types that are mostly mineral-based or synthetic leads to accidental
but unavoidable inflow of considerable quantities of non-biodegradable lubricants into the
environment. The increase in ecological concern inspires research in the lubricant industry
into raw materials from renewable sources.
Biodegradability is the ability of a substance to be decomposed by microorganisms. Veg-
etable oils have biodegradability of 97% to 99%, but that of mineral oils is only 20% to 40%,
according to Rudnick (2006). However, vegetable oils have failed to meet the demands of indus-
trial lubricants by not having acceptable physical and tribological properties. Researchers are
seeking methods and additives to improve these properties to an acceptable level so that there
can be considerable reductions in the discharge of non-biodegradable oils into the environment.
There are various methods of improving the tribological and physical properties of vegeta-
ble oils. Additives and chemical modification are the most frequently adopted methods. Vari-
ous classes of additives, such as extreme pressure additives, pour-point depressors, viscosity
modifiers, corrosion inhibitors, and nanoparticles, are used nowadays to improve such prop-
erties. One of the most commonly used additives is zinc dialkyldithiophosphate (ZDDP).
2 EXPERIMENTAL DETAILS
2.1 Equipment used

The equipment used for the experiments is described below.
417

Figure 1. Schematic of four-ball tribo tester.
2.1.1 Ducom TR-30L-PNU-IAS Four-ball tester

A four-ball tribo tester has three balls in the ball pot at the bottom that are stationary, and
one ball at the top, attached to a rotating spindle (see Figure 1). The ball pot is filled with the
oil sample to be tested. There is provision to heat the oil inside the ball pot. Load is applied
to the rotating ball as per ASTM (1999) standard during rotation for a specified time (3,600
seconds here).
During the experiment, a real-time graph of Coefficient of Friction (CoF) can be observed.
After the experiment, the stationary balls are examined under a microscope to measure the
Wear Scar Diameter (WSD).
2.1.2 Brookfield DV2T extra viscometer

Viscosity indexes were determined using a procedure based on international standard ASTM
D2270-04. In order to calculate these, the kinematic viscosity of oils at 40°C and 100°C were
experimentally measured using a viscometer.
2.2 Materials used

CoF, WSD, and viscosity tests were conducted on various oil samples as follows:
a. pure Palm Kernel Oil (PKO)
b. SAE 20W40 oil
c. PKO + ZDDP
In the latter sample, the weight percentage of ZDDP was varied between 0.5, 1.0, 1.5, 2.0, and
2.5%. CoF and WSD were determined using a Ducom four-ball tester with test conditions of 392
N load, 75 °C, 1200 RPM and 3,600 seconds. For this testing purpose, chromium alloy steel balls
as per ASTM D2783 (IP 239) were used. Three balls at the bottom were fixed in a ball pot and one
ball was attached to the motor spindle. Load is given via the output from the air compressor. The
chemical properties of the oils were analyzed as per American Oil Chemists’ Society methods.
The main challenge in using vegetable oil is the inconsistency of the properties of the
oil samples. The properties of each oil sample depend on environmental factors wherein
the mother plant grows, including soil quality, humidity, temperature, and water availability.
Because of this, test results can vary from sample to sample. Samples were sourced from three
different places; all tests were repeated on these three samples and average results taken if
the difference between maximum and minimum values was within 1%. In tests for which this
could not be achieved, worst values were taken.
3.1 Effect of ZDDP on anti-wear properties

The wear scar diameter of PKO was found to decrease with increasing concentration of
ZDDP additive up to a particular limit. PKO + 1.5% ZDDP showed the lowest value of
418

Figure 2. Coefficient of friction vs oil samples tested.
Table 1. CoF and WSD values of various oil samples tested.
WSD
Viscosity
Oil name CoF Micrometer index
PKO 0.0513 460.97 146

SAE 20W40 0.0475 498.39 122
PKO + 0.5% ZDDP 0.0424 449.33 151
PKO + 1.0% ZDDP 0.0411 421.55 156
PKO + 1.5% ZDDP 0.0406 401.37 161
PKO + 2.0% ZDDP 0.0437 429.82 163
PKO + 2.5% ZDDP 0.0496 487.94 169
WSD. Azhari et al. (2015) explained the mechanism behind this as the reaction of ZDDP
with the metal surface to form a solid protective film and the reaction layer. When metal is
immersed in ZDDP solution in a lubricant or other non-polar solvent, a thermal film rapidly
forms at the metal surface.
The WSD and CoF values of pure PKO, PKO with additives, and SAE 20W40 oil are
shown in Figure 2 and Table 1.
3.2 Effect of ZDDP on anti-friction properties

Initially, it was observed that as the concentration of ZDDP increases in PKO, the CoF
decreases. But this trend continues only up to 1.5 Wt.% of ZDDP, after which CoF increases.
According to Mahipal et al. (2014), the former observation is due to a hydrodynamic bound-
ary film being formed at this optimal concentration. At this concentration, frictional torque
on the contacting surfaces will be reduced and the coefficient of friction at the contact sur-
faces will be reduced. However, at higher concentrations, the excess ZDDP adversely affects
boundary film formation, due to excess zinc adsorption on the contact surfaces, leading to
an increase in the frictional torque. The CoF values of pure PKO and PKO with additives are
as shown in Figure 2 and Table 1.
3.3 Effect of ZDDP on chemical properties

The chemical properties of PKO are affected by ZDDP additive. It improved the saponifica-
tion value and reduced the acid value and iodine value. The saponification, iodine, acid and
419

Figure 3. Wear scar diameter (in micrometers) vs oil samples tested.
Table 2. Chemical properties of PKO and PKO + 1.5% ZDDP.
Saponification value Iodine value Acid value Ester value
Oil name mg KOH/g g/100g mg KOH/g mg KOH/g
PKO 244 21 3.2 277.8

PKO + 1.5% ZDDP 256 16 2.9 292.3
Table 3. Thermal properties of servo oil, palm kernel oil, and palm kernel oil + 1.5% ZDDP.
Flash point Fire point Pour point Cloud point
Oil name °C °C °C °C
Servo oil 230 260 -6 -3

PKO 242 251 22 23
PKO + 1.5% ZDDP 256 264 12 14
ester values of PKO and PKO + 1.5% ZDDP, which gave the best results in wear and friction
tests, are shown in Table 2.
3.4 Effect of ZDDP on thermal properties

Palm kernel oil exhibits a good flash point and fire point. The Achilles’ heel of PKO is its
melting point. Pour point and cloud point are high for a lubricant.
As the ZDDP additive has better thermal properties, as explained by Azhari et al. (2014),
it can improve the thermal properties of the oil. The thermal properties of PKO, PKO + 1.5%
ZDDP, and servo oil are shown in Table 3.
4 CONCLUSION
Tribological tests show that palm kernel oil has better anti-wear and anti-friction properties
than SAE 20W40 oil. PKO samples with ZDDP additive resulted in minimum wear scar
diameter and coefficient of friction. The viscosity index was also comparable with commer-
cial 20W40 oil. The chemical properties of palm kernel oil are better than all other vegetable
420

oils we have tested, including coconut oil, soybean oil, sunflower oil, mustard oil and karanja
oil. In terms of thermal properties, flash point and fire point are on a par with servo oil.
Properties that show poorer results are pour point and cloud point. Palm kernel oil is in a
semi-solid state at 20°C. Though the melting point value is depressed considerably after add-
ing ZDDP, PKO is still poor compared to commercial SAE 20W40 oil. According to Mahipal
et al. (2014), this can be improved by esterification, which has been proven to suppress the
melting pour point of such oils.
REFERENCES
American Society for Testing Materials (ASTM). (1999). D4172-94: Standard Test Method for Wear
Preventive Characteristics of Lubricating Fluid (Four-Ball Method). West Conshohocken, PA: ASTM
International.
Azhari, M.A., Fathe’li, M.A., Aziz, N.S.A., Nadzri, M.S.M. & Yusuf, Y. (2015). A review on addition
of zinc dialkyldithiophosphate in vegetable oil as physical properties improver. ARPN Journal of
Engineering and Applied Sciences, 10(15), 6496–6500.
Azhari, M.A., Suffian, Q.N. & Nuri, N.R.M. (2014). The effect of zinc dialkyldithiophosphate addition
to corn oil in suppression of oxidation as enhancement for bio lubricants: A review. ARPN Journal
of Engineering and Applied Sciences, 9(9), 1447–1449.
Balamurugan, K., Kanagasabapathy, N. & Mayilsamy, K. (2010). Studies on soya bean based lubricant
for diesel engines. Journal of Scientific & Industrial Research, 69, 794–797.
Barnes, A.M., Bartle, K.D. & Thibon, V.R. (2001). A review of zinc dialkyldithiophosphates (ZDDPs):
Characterisation and role in the lubricating oil. Tribology International, 34, 389–395.
Dinda, S., Patwardhan, A.V., Goud, V. & Pradhan, N.C. (2008). Epoxidation of cotton seed oil by aque-
ous hydrogen peroxide catalyzed by liquid inorganic solids. Bio Resource Technology, 99, 3737–3744.
Erhan, S.Z., Sharma, B.K. & Perez, J.M. (2006). Oxidation and low temperature stability of vegetable
oil-based lubricants. Industrial Crops and Products, 24, 292–299.
Luna, F.M.T., Cavalcante, J.B., Silva, F.O.N. & Cavalcante, C.L., Jr. (2015). Studies on biodegradability
of bio-based lubricants. Tribology International, 92, 301–306.
Mahipal, D., Krishnanunni, P., Mohammed Rafeekh, P. & Jayadas N.H. (2014). Analysis of lubrication
properties of zinc-dialkyl-dithio-phosphate (ZDDP) additive on karanja oil (Pongamia pinnatta) as
a green lubricant. International Journal of Engineering Research, 3(8), 494–496.
Rudnick, L.R. (Ed.). (2006). Synthetics, mineral oils, and bio-based lubricants: Chemistry and technology.
Boca Raton, FL: CRC Press.
Zulkifli, N.W.M., Azman, S.S.N., Kalam, M.A., Masjuki, H.H., Yunus, R. & Gulzar, M. (2014). Lubric-
ity of bio-based lubricant derived from different chemically modified fatty acid methyl ester. Tribol-
ogy International, 93, 555–562.
421

Influence of perforated tabs on subsonic set control
Dharmahinder Singh Chand & D.L. Vasthadu Vasu Kannah

Department of Aeronautical Engineering, Tagore Engineering College, Chennai, India
S. Thanigaiarasu
Department of Aerospace Engineering, MIT Campus, Anna University, Chennai, India
S. Elangovan
Department of Aerospace Engineering, Bharath University, Chennai, India
ABSTRACT: An experimental analysis has been carried out to examine the mixing encour-
aging effectiveness of tabs with circular perforation of different perforation diameters,
1.5 mm, 2 mm and 2.5 mm. The geometrical blockage offered by the perforated tabs, placed
diametrically opposite, at the nozzle exit, were 8.42%, 7.55% and 6.42%, respectively, for per-
foration diameters 1.5 mm, 2 mm and 2.5 mm. The Mach along the jet central axis and Mach
profiles in the directions along and tangential to the tabs were calculated at various axial
locations. The results of the Mach 0.4, 0.6 and 0.8 jets studied show that the tab with 2 mm
perforation is a better mixing promoter than 1.5 mm and 2.5 mm perforations, effecting in a
core length decrement of 62% for Mach 0.6 jet. The corresponding reduction in core length
for 1.5 mm and 2.5 mm holes are only 47.61% and 42% respectively.
Keywords: Perforated tab; passive control; potential core; mixing enhancement
NOMENCLATURE
X = Co-ordinate jet central axis

Y = Co-ordinate along the tabs
Z = Co-ordinate tangential to the tabs
NPR = Nozzle Pressure Ratio
ϕp = Diameter of perforation
Mj = Jet Mach number
M = Localized Mach number
P = Pitot measured pressure
P0 = Settling chamber pressure
1 INTRODUCTION
Restricting of jet has become an active area of analysis owing to its application potential,
such as improvement of stealth capabilities, minimization of base heating and reduction of
aero acoustic,. Most of these applications require mixing improvement of the jet, i.e. the
mass from the adjoining region entrained by the jet has to be mixed with the jet fluid mass as
rapid as possible (Reeder & Zaman, 1996).
To achieve mixing enhancement, mixing promoting small scale vortices needs to be origi-
nate and commence at the nozzle exit (Reeder & Samimy, 1996), with this aim considerable
number of passive and active techniques have been identified by the researchers over the
recent few decades (Rathakrishnan, 2010). Among these, passive control in the form of tab
423

has become popular due to its simple geometry and efficient mixing, promoting capability
(Chiranjeevi. & Rathakrishnan, 2010).
The effect of wedge-shaped protrusions on the progress of a low-speed jet was examined
by Bradbury and Khadem (1975). This tab was helpful to distort the jet propagation. Bohl
and Foss (1996), Wishart et al. (1993) and Zaman et al. (1994) have determined that the
tab creates a pair of counter-rotating stream-wise vortices. Zaman (1993) reported that the
relative volume of the crest stream-wise vortices was observed to be about 20% of that of
the crest azimuthal vortices for a tabbed circular jet at a Mach number of 0.3. Zaman et al.
(1992) established that the distortion introduced by a mechanical tab is due to combination
of duo stream-wise vortices. These vortices were found to be accountable for the phenomenal
entrainment. Bohl and Foss (1996) reported that the two achievable sources of stream-wise
vortices for the flow above a tab are the pressure gradients, which flux stream wise vortices
into the flow and the well-known ‘necklace’ or ‘horseshoe’ vortices owing to boundary layer
reorientation.
A rapid development of the mixing layer was observed by Reeder and Samimy (1996)
when tabs were inserted. The consequence of the stream-wise tab position within the nozzle
was investigated by Reeder and Zaman (1996). They observed that tabs placed at the exit
plane increased entrainment into the jet, whereas tabs located further upstream in the nozzle
caused an ejection of core fluid. Sreejith and Rathakrishnan (2002) studied a wire across a
diameter (cross-wire) as a passive control to improve the jet mixing. The stream-wise eddied
inducted by the cross-wire lead to an additional and rapid decay of the centerline pitot pres-
sure. Also, it was noticed that the cross-wire deteriorated the shocks in the jet core signifi-
cantly. The authors confirmed that the limit for tab length is the nozzle exit radius and not
the boundary layer thickness. This limit of tab length is termed as Rathakrishnan limit.
Vortices emitted from the edges of flat and arc plates, kept normal to a consistent water
flow, were investigated by Takama et al. (2008). It was found that the lookalike vortex in
the rear of plate was much bigger than that for a circular cylinder. This indicated the major
role played by the reverse flow on the vortex formation. Extending this effect Thanigaiarasu
et al. (2013) studied the effect of arc tabs on the mixing behavior of subsonic and correctly
expanded sonic jets. Two tabs in the form of a semi-circular arc of diameter 1.5 mm and
length 2 mm were placed at the nozzle exit of 10 mm in diameter. The mixing in close prox-
imity of the jet was studied for two designs of the tab; namely concave surface of the arc tab
in front of the flow exiting the nozzle (arc tab facing in) and the convex surface of the front
of the flow (arc tab facing out). Mixing promotion caused by these tabs was evaluated with a
plain rectangular tab of the same blockage ratio of 7.64%. It was discover that a core length
reduction of 80% was achieved in arc tab facing in, whereas the corresponding reduction in
core length with arc tab facing out was just 50%. Mach profiles in the planes tangential to the
jet axis showed that the arc tab facing in deforms the jet effectively by forcing the jet to spread
faster and wider in the plane normal to the tab than that along the tab.
The proficiency of corrugated tabs for high speed jet control was studied by Chiranjeevi
and Rathakrishnan (2010). They employed two tabs of rectangular shape with a 4.2% block-
age; the edges were corrugated and located diametrically opposite at the convergent-divergent
nozzle exit of a Mach 1.8. It was observed that a better mixing efficiency was achieved com-
pared with the uncontrolled jet. More than 78% reduction in core length was obtained with
corrugated tabs for the jet operated at a nozzle pressure ratio of 7; the corresponding reduc-
tion with plain tab was noticed at only 54%. It was found that the mixing capability of the
corrugated tabs improved consistently with increase in nozzle pressure ratio.
Arun Kumar and Rathakrishnan (2013a) studied the influence of triangular tabs on super-
sonic jets. They demonstrated superiority of triangular tabs in promoting jet mixing in all
three jet zones with Mach 2, in the presence of slightly favorable and unfavorable pressure
gradients. Also, they investigated triangular corrugated tabs for jet control. It was demon-
strated that among triangular tabs, the tab with a truncated apex is found to be a better mix-
ing enhancer than the one with a sharp apex. The reason for the superiority is envisaged as
the fact that the vortices at the tab tip do not interrelate themselves and drop energy, as in the
case of the pointed apex. Corrugations at the edges of the tabs greatly augment the jet mixing.
424

Truncated corrugated triangular tabs were also found to be effective in boosting the mixing
capability by Arun Kumar and Rathakrishnan (2013b). Semi-circular as well as square cor-
rugated tabs were beneficial to bifurcate the jet into two fractions (lobes) at X/D ≤ 1. Also
they established that the shape of corrugation plays a major role in the mixing enhancement.
Tab geometries, such as rectangular, triangular, circular, a d arc have been studied extensively.
However, these solid tabs shed mixing, encouraging small eddies of homogeneous size only
from a specified height of the tab. But it may be more beneficial from a mixing point of view,
if the vortices themselves possess varied sizes. To generate such small vortices of varied size,
tabs with perforation over their flat faces are explored in the present investigation.
2 EXPERIMENTAL DETAILS
The experiments were carried out in the High Speed Jet Laboratory at MIT, Anna University,
Chennai. The test facility consists of an air delivering system (compressors and reservoir)
and an open jet testing facility as shown in Figure 1.
2.1 Open jet

Exit diameter (D) 20 mm of convergent nozzle prepared from gunmetal was used in this
study. The nozzle was fixed at the end of the settling chamber with an o-ring sealing to
minimize the leakage. Required stagnation pressures were protected in the settling chamber
to generate the desired subsonic jet Mach number. A pitot tube made of stainless steel tube
of inside diameter 0.4 mm and outside diameter 0.6 mm, fixed on a traverse mechanism, was
used for pressure (P) measurements along and across the jets. The pitot probe has to be small
enough such that there is insignificant disturbance to the incoming flow because of its exist-
ence. Thus the proportion of nozzle exit area to the probe area is (10/0.6)2 = 1111.11, which is
higher than the limit of 64, for concerning the probe blockage as minor4. The pitot tube was
linked to a pressure transducer for pressure measurements.
2.2 Tab jets

Perforated tabs of stainless steel were used in the present study. The length and breadth of
the tab were 5 mm and 3 mm, respectively. Two identical tabs were placed, at diametrically
opposite positions, at the nozzle exit as shown in Figure 2. The fraction of the tab area nor-
mal to the flow nozzle exit area studied were 8.42%, 7.55% and 6.42%. The pitot pressures
measured were translated into the Mach number using isentropic relations, using the ambient
pressure in the laboratory as the static pressure, because subsonic jets are always considered
to be correctly expanded.
Figure 1. An artistic view of high speed laboratory.

1. 80 HP induction motor, 2. Reciprocating compressors, 3. Activated charcoal filter, 4. Water cooling
unit and silica gel dryer units, 5. Storage tanks, 6. Gate valve, 7. Pressure regulating valve, 8. Settling
chamber, 9. Traversing system, 10. Instrumentation desk.
425

3.1 Centerline Mach number

Centerline jet decay can be taken as a genuine measure of promoting jet mixing (Rathakrishnan,
2009). The centerline jet reduction of uncontrolled and controlled Mach 0.4 are compared in
Figure 5. It is seen that the potential core span of the jet is reduced by 58.33% with the tab
with 7.55% blockage (2 mm circular perforation), while reduction is 47.91% for the tab with
8.42% blockage (1.5 mm circular perforation) and 42.7% for the tab with 6.42% blockage
(2.5 mm circular perforation) compared with uncontrolled jet. It may be due to non-uniform
Figure 2. A pictorial view of tab arrangement at nozzle exit.
Figure 3. Schematic diagram of tab at nozzle exit and co–ordinates implemented.
Figure 4. Physical mechanism behind vortex generation between perforation and the tab.
426

Figure 5. Centerline Mach decay at Mach 0.4 jet.
vortices emitted by web and perforation. Also it is observed that decay in transition region is
very rapid beyond 11D compared with uncontrolled jet. This is due to the fact that small scale
vortices are stronger than large scale vortices and travels longer distance 20D.
Figure 6 depicts the variation in the potential core area at Mach 0.6 for uncontrolled jet,
tabs with 8.42% blockage (1.5 mm circular perforation), 7.55% blockage (2 mm circular per-
foration) and 6.42% blockage (2.5 mm circular perforation). Potential core for uncontrolled
jet extends almost up to X/D = 5.25. The core span extends to X/D = 2 for the tab with
7.55% blockage (2 mm circular perforation), whereas core length extends to X/D = 2.75 and
X/D = 2.78 for the tab arrangements of 8.42% blockage (1.5 mm circular perforation) and
6.42% blockage (2.5 mm circular perforation), respectively. The tab with 7.55% blockage
(2 mm circular perforation) is more effectual for mixing enrichment than other perforation
tab configurations studied.
This perhaps may be because of the varying radii of curvature created by perforation inside
the tab, which is responsible for manipulating the dimensions of vortices (Quinn 1995).
427

It is apparent from Figure 7 that the potential core span is reduced by 50% with the tab
with 7.55% blockage (2 mm circular perforation) and the core reduction with tabs of 8.42%
blockage (1.5 mm circular perforation) and 6.42% blockage (2.5 mm circular perforation) is
40% and 38% respectively, compared with uncontrolled jet. The jet stirring efficiency with
atmospheric air is the maximum for the tab with blockage of 7.55% (2 mm diameter circular
perforation) among other perforated tabs because of ‘necklace’ or ‘horseshoe’ vortices due to
the boundary layer rearrangement over a tab (Chiranjeevi & Rathakrishnan, 2010).
When the perforation diameter is 1.5 mm the mass, or more precisely energy, of the jet
through the perforation is less, resulting in lower shear layer interaction between core jet
and perforated jet. Also the eddies liberated by the perforation and web are of the same size
as in Figure 4 (a). This combination results in lesser mixing enhancement. Similarly energy
through the perforation of diameter 2.5 mm as in Figure 4 (c) is greater compared with
1.5 mm and 2 mm perforated tabs. But the combination of shear layer interaction (between
core jet and perforation jet) and vortices exited by the web (w) (shortest distance between the
outer edge of tab and inner edge of the perforation) to promote mixing are insufficient. The
reason may be due tp the reduced web width owing to the increased perforation diameter.
3.2 Mach number profile

Figure 8 shows that more dip and sudden rise in regard to tabs with 7.55% blockage (2 mm cir-
cular perforation) due to combination of vortices and perforation at X/D = 0.15 for Mj = 0.4,
which in turn entrains mass from atmosphere to promote mixing efficiently. This may be due
to the fact that non-circular jets entrain more ambient fluid than their circular counterparts,
and also to the non-uniform self-induction brought about by azimuthal curvature variation
of the initial vortices generated at the nozzle exit plane (Mrinal et al., 2006).
Figure 8 shows the shrunk area compared with other controlled and uncontrolled flows
studied for the outcome of perforation size as an indication of the scattering of vortices
along the tab applicable to the tab with 7.55% blockage (2 mm circular perforation) at axial
location of X/D = 1 for jet Mach 0.4.
Figure 9 exhibits that for all perforated tabs, Mach number variation in the perpendicular
plane is almost identical to the uncontrolled jet at X/D = 0.15 for Mj = 0.4. This implies that
the thrust loss caused by the tabs is only marginal7. Also from Figure 11 it is clear that the
existence of core is observed up to Z/D = 0.4, for the tab with 7.55% blockage (2 mm circular
428

Figure 8. Y-Mach number profiles at X/D = 0.15 for Mj = 0.4.
Figure 9. Y-Mach number profiles at X/D = 1 for Mj = 0.4.
Figure 10. Z-Mach number profiles at X/D = 0.15 for Mj = 0.4.
429

Figure 11. Y-Mach number profiles at X/D = 0.15 for Mj = 0.8.
Figure 12. Z-Mach number profiles at X/D = 0.15 for Mj = 0.8.
perforation), whereas it is about Z/D = 0.7 and Z/D = 0.6 for tabs with 8.42% blockage
(1.5 mm circular perforation) and 6.42% blockage (2.5 mm circular perforation) respectively,
at axial position X/D = 0.15 and 0.25. This shows that the tab with 7.55% blockage (2 mm cir-
cular perforation) alters jet cross-sectional area more effectively and entrains more air from
the nearby compared to other perforated tabs studied.
From Figure 10, it is perceived that the mixing enhancement performance in XY-plane is
better for tab with 7.55% blockage (2 mm circular perforation) due to more shrunk in area.
Figure 12 proves that the jet develops faster along Z/D for tab with 7.55% blockage (2 mm
circular perforation) at X/D = 0.15, for Mach 0.8. This may be connected to the drift of small
eddies sidewise from the nozzle outlet.
3.3 Iso-mach contour

The pitot pressure variations in the jet field with control is measured in the YZ-plane and
converted into Mach number using isentropic relation at axial locations, as illustrated in
Figure 13. Three pairs of vortices can be noticed from the Figure 13 (b), each at root, tip and
web. This may be credited to suitable combination of web and perforation for shedding vary-
ing sizes of vortices. In the case of 1.5 mm perforation only two pairs can be seen from Fig-
ure 13 (a). Again from Figure 13 (c) a pair of vortices is visible at root only. This means that
the tab with 7.55% blockage (2 mm circular perforation) is more capable mixing promoter
among other circular perforation tabs, namely, 8.42% blockage (1.5 mm circular perforation)
and 6.42% blockage (2.5 mm circular perforation). Also there are dual pressure mountains
on both sides of the jet axis, as seen in Figure 13 (b). It may be taken as an indication of jet
bifurcation13, which is a vital factor for mixing enhancement.
430

Figure 13. Iso-Mach contour for tabs, at X/D = 0.1, for Mj = 0.6. (a) 1.5 mm circular perforation (b)
2 mm circular perforation (c) 2.5 mm circular perforation.
4 CONCLUSION
The outcome of the current study on subsonic jet control with perforated tabs show that all
the perforations studied are profound and effective in promoting mixing close to jet field.
However, among them, perforation of 2 mm diameter is a highly effective stirring promoter,
resulting in a core span reduction of 62%. The corresponding core span reduction for 1.5 mm
and 2.5 mm holes are only around 47%.
REFERENCES
Arun Kumar, P. & Rathakrishnan, E. (2013a). Corrugated triangular tabs for supersonic jet control.
Journal of Aerospace Engineering, 1–15.
Arun Kumar, P. & Rathakrishnan, E. (2013b). Corrugated truncated triangular tabs for supersonic jet
control. Journal of Fluid Mechanics, 135, 1–11.
Arun Kumar, P. & Rathakrishnan, E. (2013c). Truncated triangular tabs for supersonic—jet control.
Journal of Propulsive Power, 29, 50–65.
Bohl, D. & Foss, J.F. (1996). Enhancement of passive mixing tabs by the addition of secondary tabs
AIAA paper, 96–054.
Bradbury, L.J.S. & Khadem, A.H. (1975). The distortion of a jet by tabs. Journal of Fluid Mechanics,
70 (4), 801–813.
431

Chiranjeevi, P.B. & Rathakrishnan, E. (2010). Corrugated tabs for supersonic jet control. Journal of
AIAA, 48(2), 453–465.
Elangovan, S. Solaiappan, A. & Rathakrishnan, E. (1996). Studies on twin elliptic jets. Journal of Aero-
space, 100(997), 295–296.
Mrinal, K., Pankaj, S.T. & Rathakrishnan, E. (2006). Studies on the effect of notches on circular sonic
jet mixing. Journal of Propulsive Power, 22, 211–214.
Quinn, W.R. (1995). Turbulent mixing in a free jet issuing from a low aspect ratio contoured rectangular
nozzle. Journal of Aerospace, 99, (988), 337–342.
Rathakrishnan, E. (2009). Experimental studies on the limiting tab. Journal of AIAA, 47(10),
2475–2485.
Rathakrishnan, E. (2010). Applied gas dynamics. New Jersey: Wiley.
Reeder, M.F. & Samimy, M. (1996). The evolution of a jet with vortex-generating tabs: real-time visuali-
zation and quantitative measurements. Journal of Fluid Mechanics 311, 73–118.
Reeder, M.F. & Zaman, K.B.M.Q. (1996). Impact of tab location relative to the nozzle exit on jet distor-
tion. Journal of AIAA, 34(1), 197–199.
Sreejith, R.B. & Rathakrishnan, E. (2002). Cross-wire as passive device for supersonic jet control. AIAA
paper, 2002–4052.
Takama, Y., Suzuki, K. & Rathakrishnan, E. (2008). Visualization and size measurement of vortex shed
by flat and arc plates in a uniform flow. International Journal of Review of Aerospace Engineering,
1(1), 55–60.
Thanigaiarasu, S., Jayaprakash, S., Elangovan, S. & Rathakrishnan, E. (2008). Influence of tab geome-
try and its orientation on under expanded sonic jets. Journal of Aerospace Engineering, 222, 331–339.
Wishart, D.P., Krothapalli, A. & Mungal, M.G. (1993). Supersonic jet control disturbances inside the
nozzle. Journal of AIAA, 31(7), 1340–1341.
Zaman, K.B.M.Q. (1993). Streamwise vorticity generation and mixing enhancement in free jets by
delta-tabs. AIAA paper, 93–3253.
Zaman, K.B.M.Q., Reeder, M.F. & Samimy, M. (1992). Supersonic jet mixing enhancement by delta-
tabs. AIAA paper, 92–3548.
Zaman, K.B.M.Q., Reeder, M.F. & Samimy, M. (1994). Control of an axi-symmetric jet using vortex
generators. Physics of Fluids, 6, 778–793.
432

Renewable energy and alternate fuels

Modification of a parabolic trough collector and its exergy

analysis
O. Arjun, M.C. Nikhil & B. Sreejith

Government Engineering College, Kozhikode, Kerala, India
ABSTRACT: Extraction of thermal energy from solar energy is a challenge in harvesting

solar energy. When higher temperatures are required, it becomes necessary to concentrate
solar radiation. Parabolic Trough Collectors (PTCs) are formed by a cylindrical surface of
mirrors having high reflectivity with a parabolic shape that concentrates solar radiation on a
receiver tube located at the focal point of the parabola. Working fluid, which is heated up by
the solar radiation, circulates inside the receiver tubes. Energy losses at higher temperatures
are the main cause of efficiency loss in PTCs. In previous work, the collector was exposed to
the atmosphere and thus thermal losses were high. Some modifications to the design of PTCs
for reduction of energy losses was proposed in this work. In order to reduce convective losses,
an evacuated tube collector was introduced instead of copper tube. A modified PTC was
fabricated and its performance evaluated. Exergy analysis was performed to study the effects
of operational and environmental parameters on the performance of PTCs.
Keywords: Parabolic Trough Collectors, Exergy analysis, solar radiation
1 INTRODUCTION
The sun is the head of the family of planets, and is the most abundant source of renewable
energy for our earth. Reserves of other energy sources, such as coal and fossil fuel, will even-
tually diminish. Solar energy contains radiant heat and light energy from the sun, which can
be harnessed with modern technologies like photovoltaic (PV) cells, solar heating, artificial
photosynthesis, and solar thermal electricity. Solar thermal collectors gain energy through
radiation, conduction and convection. A flat plate collector will lose energy through conduc-
tion as well as convection; thus, it reduces the amount of energy that can be transferred to
working fluid. The evacuated tube collector is a new harnessing technology. It is ideal for
high-temperature applications such as boiling water, pre-heating, and steam production.
Exergy is the ability of a system to do useful work before it has been brought into thermal,
mechanical and chemical equilibrium with the environment. It is derived from both the first
and second laws of thermodynamics. When a system and its surroundings are not in equilib-
rium with each other, then we can extract work. This means that if there is any difference in
temperature between a system and its surroundings, it will be in unstable equilibrium. This
situation can be used to produce work. On the basis of the second law of thermodynamics, it
is impossible to convert low-grade energy completely into shaft work. The part of low-grade
energy that can be converted into useful work is termed available energy or exergy. The per-
formance of Parabolic Trough Collectors (PTCs) can be explained in terms of exergy, which
provides a useful basis for the design and optimization of PTCs.
The present work includes modification of an existing PTC with substitution of an evacuated
tube. The main drawback of the existing system is that the outside surface of the receiver tube
is exposed to the atmosphere; thus convective energy loss will be dominant, which reduces the
performance of the PTC. In order to reduce this loss, evacuated tubes are introduced instead of
copper tubes. Exergy analysis was performed to assess the performance of the modified PTC.
435

Yadav et al. (2013) conducted an experiment on parabolic troughs with various reflectors,
such as stainless-steel sheet, aluminum foil, and aluminum sheet, and obtained temperatures
of 42.1°C, 48.2°C and 52.3°C, respectively. A buffed aluminum sheet will produce even better
results in comparison to an untreated sheet.
Previous work by Nikhil and Sreejith (2016) used acrylic glass and its performance using
Al2O3 nanofluid as the working fluid was studied. The main limitation of the model is that
the collector is exposed to ambient air and thus heat loss to the surroundings is higher. In
the current work, the collector is replaced with an evacuated tube collector and an exergy
analysis is performed.
Jafarkazemi and Ahmadifard (2013) conducted an energy and exergetic analysis of a flat
plate collector; based on their theoretical results, the maximum energy and exergy efficiency
of a flat plate collector are close to 80% and 8%, respectively. Energy efficiency is always
higher than that of exergy efficiency due to irreversible work associated with flow.
A study was performed by Padilla et al. (2014) to simulate the heat transfer model for a
PTC. They made an exergy balance of the receiver in the control volume. They also calculated
exergy efficiency and destruction rates.
Tyagi et al. (2007) performed a parametric study of concentrating-type solar collectors for
different mass flow rates and concentration ratios using hourly solar radiation. They con-
cluded that exergy output and thermal efficiency are an increasing function of the mass flow
rate for a given value of solar intensity.
The performance and efficiency of PTCs have increased significantly during the last three
decades. This is due to the development of new technologies and advances in material sci-
ence. New collectors, such as evacuated tubes and Apercus evacuated tubes, can absorb solar
energy more effectively than flat plate collectors. However, combining these collectors in a
concentrating technology like a parabolic trough, with dishes and Fresnel lenses will enhance
the overall performance of the system.
2 THERMAL PERFORMANCE OF PTC
The performance of a PTC can be estimated by the efficiency factor η, which is defined as
the ratio of the net heat gain to the solar radiation energy, based on the diffuse reflection area
of the solar collector:
p (Tout − Tin )
mC
η=
I ( A a − As ) ρ
Calculation of the thermal performance of concentrating collectors is similar to that of

flat plate collectors. Because concentrating collectors operate at higher temperatures than flat
plate collectors, thermal loss will be greater. In order to calculate thermal loss, we must know
the overall heat loss coefficient, UL, which is given by:
Q loss
UL = (2)
A r (Tr − Ta )
2.1 Design of parabolic trough

In the present study, the material used for the PTC reflector is acrylic glass, whose reflectiv-
ity is nearly the same as MIRO-SUN® (Alanod GmbH, Ennepetal, Germany), which is used
in reflectors in concentrating collectors. The dimensions and specification of the parabolic
trough are shown in Figure 1 and Table 1, respectively.
2.2 Exergy analysis

The exergy balance equation must be applied on the control volume of the receiver. Accord-
ing to Padilla et al. (2014), the partial differential equation of the exergy balance is given by:
436

Figure 1. Dimensions of the trough.
Table 1. Specifications of PTC.
Parameter Value
Width of the concentrator, W 1 m

Length of the concentrator, L 1.8 m
Rim angle of the trough, φ 90°
Concentration ratio, C 5.2
Aperture area, Aa 1.8 m2
Focal length 0.25 m
Half acceptance angle 1.5°
dEcv

dt
= ∑ E
j
qj − W cv + ∑ m i e fi − ∑ m e e fe − E d − E loss
i e
It is assumed that steady state flow, and kinetic and potential energy were negligible, and
that the specific heat and other properties remain constant during operation.
Exergy efficiency is defined as the ratio of exergy gain to maximum possible solar radia-
tion exergy, that is:
E E − E
ηex = gain = i e
E sr E sr
 Te Te C ( T ) 
m  ∫ C p ( T) dT − To ∫ p
dT 
Ti Ti T
ηex =   (3)
I b ( A a − As )ρψ
 T
m C p (Te − Ti ) − To ln e 
 Ti 
ηex =
I b ( A a − As )ρψ
where ψ is the relative potential of the maximum useful work extracted from radiation and is
calculated with Petela’s (2003) formula:
4
4T  1T 
ψ = 1−  o  +  o 
3  Ts  3  Ts 
3 EXPERIMENTAL FACILITY
In order to avoid confusion in terminology and ensure consistency between terms we use
‘concentrating collector’ to represent the entire setup. The term ‘concentrator’ is used for the
optical subsystem that directs the solar radiation onto the absorber, and the term ‘receiver’
represents the subsystem consisting of the absorber, coating and cover, as shown in Figure 2.
437

Figure 2. Schematic diagram of PTC.
Figure 3. Experimental setup.

The experiment was conducted for six hours per day. Figure 3 shows the experimental setup
of the PTC. It consists of a storage tank, inlet pipe, outlet pipe, K-type thermocouple, tem-
perature scanner, parabolic trough, lux meter, measuring jar and anemometer. The pipes and
all fittings are made from Chlorinated Polyvinyl Chloride (CPVC), which is used in solar
heating and high-temperature applications instead of ordinary PVC because of its ability to
withstand high temperatures. A lux meter and anemometer are used to measure solar inten-
sity and wind velocity, respectively, at a particular instant. The flow rate was measured with a
measuring jar in which the time for a 200 ml rise was taken as the reference. Thus, we obtain
the mass flow rate in kg/s. Minerally insulated K-type (nickel-chromium/nickel-alumel) ther-
mocouples are used in this work. The thermocouples and all the readings of temperature
described here are calibrated. The storage tank is placed above the receiver so that water
will flow using gravitational force, which eliminates a pumping system. There are two valves,
located at the inlet and outlet of the receiver; thus, we can control the flow rate.
The experiment was carried out during February to March 2017 at the Government Engi-
neering College, Kozhikode (11.2858° N, 75.7703° E), between the hours of 9 a.m. and
3 p.m. The experiment was conducted in differing climatic situations, such as cloudy and
sunny skies. Further, the study was carried out in different solar intensities, varying from 600
to 1000 W/m2. The efficiency of the PTC was calculated by using Equation 1.
438

Figures 4 and 5 represent the variation of outlet temperature and efficiency with respect to
mass flow rate. As the mass flow rate increases, the outlet temperature is reduced. That is, for
lower mass flow rates the outlet temperature was high and vice versa. In addition, when the
mass flow rate increases the heating efficiency increases.
If the mass flow rate remains fixed, the effect of solar intensity on outlet temperature can
be studied. Solar intensity is varied from 600 to 1000 W/m2. It can be clearly seen from Fig-
ure 6 that even under cloudy conditions, or when the intensity of radiation is at a minimum
(600 W/m2), we can obtain temperatures between 70°C and 92°C depending upon the mass
flow rate. Thus, the PTC is a promising technology for future energy needs. Furthermore,
because of the wide range of outlet temperature, it can be used in a wide variety of applica-
tions, such as process heating, desalination, and pasteurization.
As illustrated in Figure 7, the maximum outlet temperature was obtained between 12 noon
and 2 p.m. This is due to the fact that solar radiation falls perpendicularly on the trough at
this time and almost all radiation is collected via the entire length of the receiver. Further-
more, the higher the intensity of radiation, the higher the outlet temperature. On the other
hand, efficiency steadily decreases until 2 p.m. This is because of temperature rise as solar
intensity rises. Then efficiency increases because the high outlet temperature drops as com-
pared to reduction in intensity of solar radiation.
The overall heat loss coefficient, UL, can be computed using Equation 2. The variation in
heat loss coefficient with receiver temperature is shown in Figure 8. One can clearly observe
that as the temperature increases the heat loss coefficient also increases. This is due to the fact
that radiation loss will be higher at higher temperatures.
4.1 Comparison with previous work

The variation in efficiency with mass flow rate for different receivers, such as evacuated tube,
copper tube and aluminum tube, is shown in Figure 9. The efficiency of the PTC can be
Figure 4. Variation of outlet temperature with mass flow rate (for constant 1000 W/m2 solar
intensity).
Figure 5. Variation of efficiency with mass flow rate (for constant 1000 W/m2 solar intensity).
439

Figure 6. Variation of outlet temperature with solar intensity (for fixed mass flow rate).
Figure 7. Variation of outlet temperature and efficiency with time (for constant m
= 0.0025 kg/s ) .
Figure 8. Variation of heat loss coefficient with receiver temperature.
calculated using Equation 1. The evacuated tube collector shows the greatest efficiency,
nearly twice that of the copper tube collector with Al2O3 nanofluid as the working fluid. The
maximum efficiencies obtained were 58.3%, 29.3%, 18.3%, and 11% for evacuated tube, cop-
per tube with nanofluid, copper tube, and aluminum tube, respectively.
440

Figure 9. Variation of efficiency with mass flow rate.
Figure 10. Variation of exergy efficiency with mass flow rate.
4.2 Exergy efficiency

Figure 10 depicts the variation in exergy efficiency with mass flow rate. Exergy efficiency
can be calculated by using Equation 3. The variation in exergy efficiency was similar to that
of energy efficiency. However, energy efficiency will always be higher than that of exergy
efficiency. According to the second law of thermodynamics, we cannot convert low-grade
energy completely into high-grade energy because of the heat transfer irreversibility between
a system and its surroundings. This is because of the irreversible work associated with any
real process. The exergy losses include exergy flowing out to the surroundings, but in exergy
destruction indicates the loss of exergy within the system boundary due to its irreversibility.
5 CONCLUSIONS
Some modifications were performed in the design of a PTC and experiments were conducted
with different mass flow rates, various climates and different solar intensities to compare its
performance. The study demonstrated the effectiveness of the proposed modification to the
design of the PTC. Exergy analysis was performed in order to establish various parameters
that affect the performance of the PTC. This was a cost-effective way of obtaining high tem-
peratures in response to future energy demands. The following conclusions were drawn upon
the completion of the work:
• The maximum temperature obtained was 143°C, which is 33°C (∼30%) higher than that of
the previous model using copper tube with water/Al2O3 nanoparticles.
• The maximum temperature obtained for the previous model without using nanoparticles
was 96°C, which is 49% lower than that of the new model.
441

• The new model has nearly twice the efficiency compared to the previous one.
• Solar intensity plays a significant role in the performance of the PTC. At higher solar
intensity, the outlet temperature will be higher, thus giving higher efficiency.
• As the mass flow rate increases, efficiency increases and outlet temperature decreases.
• Efficiency will decrease at higher temperatures. This is because as the temperature increases
the radiative heat loss will also increase.
• The results show that exergy efficiency was much lower than energy efficiency because of
the irreversible work associated with a real process.
NOMENCLATURE
Aa Aperture area m2
As Shaded area m2
C Concentration ratio
D Diameter m
UL Overall heat loss coefficient W/m2 K
W Aperture m
E e Exergy output W
E i Exergy input W
m Mass flow rate kg/s
η Energy efficiency
ηex Exergy efficiency
θα Acceptance angle degree
ρ Reflectivity
σ Stefan–Boltzmann constant W/m2/K4
Subscripts
a Atmosphere
r Receiver
REFERENCES
Jafarkazemi, F. & Ahmadifard, E. (2013). Energetic and exergetic evaluation of flat plate solar collec-
tors. Renewable Energy, 56, 55–63.
Nikhil, M.C. & Sreejith, B. (2016). Performance evaluation of a modified parabolic trough concentrator
using nanofluid as the working fluid. NET, 192–201.
Padilla, R.V., Fontalvo, A., Demirkaya, G., Martinez, A. & Quiroga, A.G. (2014). Exergy analysis of
parabolic trough solar receiver. Applied Thermal Engineering, 67, 579–586.
Petela, R. (2003). Exergy of undiluted thermal radiation. Solar Energy, 74, 469–488.
Tyagi, S.K., Wang, S., Singhal, M.K., Kaushik, S.C. & Park, S.R. (2007). Exergy analysis and para-
metric study of concentrating type solar collectors. International Journal of Thermal Sciences, 46,
1304–1310.
Yadav, A., Kumar, M. & Balram. (2013). Experimental study and analysis of parabolic trough collector
with various reflectors. International Journal of Energy and Power Engineering, 7(12), 1659–1663.
442

Analytical computation of GWP, ODP, RF number and TEWI

analysis of various R134a/R1270/R290 blends as R22 alternatives
Sharmas Vali Shaik & T.P. Ashok Babu

NIT Karnataka, Surathkal, Karnataka, India
ABSTRACT: The principal objective of the present work is to compute the GWP, ODP,
RF number and TEWI analysis of various ternary R134a/R1270/R290 blends as alternatives
to R22. In this study thirteen refrigerant blends consists of R134a, R1270 and R290 at dif-
ferent compositions are taken. GWP and ODP of refrigerant blends are computed by using
various simple correlations. The estimation of emission of greenhouse gases and flammabil-
ity study of refrigerants are done by using TEWI and RF analysis respectively. Analytical
results revealed that all the thirteen studied fluids are ozone friendly in nature. The GWP of
refrigerant M6 (651) is lower than that of GWP of R22 (1760). RF analysis exhibited that
all the thirteen refrigerant blends are categorized as ASHRAE A2 flammability category.
Thermodynamic analysis revealed that COP of M6 (3.608) is higher that of COP of R22
(3.534). TEWI of M6 is lower among the R22 and thirteen studied fluids. Hence refrigerant
M6 (R134a/R1270/R290 50/5/45 by mass %) is an alternative to R22.
Keywords: COP, GWP, ODP, RF number, R22 alternatives, TEWI
NOMENCLATURE
HFCs Hydrofluorocarbons
HOC Heat of combustion (kJ/mol)
L Lower flammability limit (kg/m3)
MW Molecular weight (kg/kmol)
RF Refrigerant flammability (kJ/g)
U Upper flammability limit (kg/m3)
Ci Composition of ith component
m Mass flow rate of refrigerant (kg/min)
mi Mass fraction of ith component
WC Compressor work (kJ/kg)
WCP Compressor power (kW)
1 INTRODUCTION
Refrigerant R22 has adverse environmental impacts like high ozone depletion potential
(ODP) and high global warming potential (GWP) (Mohanraj et al. 2009). Therefore an
international Montreal protocol decided to phase out R22 by the year 2030 (UNEP 1987).
Currently global warming has become very significant issue and hence Kyoto protocol was
recommended to resolve this problem, for which hydrofluorocarbons (HFCs) were classified
as one among the targeted global warming refrigerants (GECR 1997). Hence in this study
an attempt was made to develop the refrigerants which will meet the requirements of both
the Montreal and Kyoto protocol respectively. Formerly various performance studies were
carried out to find suitable alternative for the refrigerant R22. Theoretical analysis revealed
that R444B was a suitable candidate to replace R22 (Atilla G.D. & Vedat O. 2015). Experi-
mental studies reported that R134a requires a larger size of compressor in order to replace
443

R22 (Devotta et al. 2001). The ODP and GWP values of various pure refrigerants were
found from the literature (ASHRAE 2009, IPCC 2014). Lorentzen suggested that R290 was
an appropriate substitute to R22 (Lorentzen G. 1995). Experimental studies exhibited that
R410A was an alternative refrigerant to replace R22 (Chen 2008). Theoretical performance
investigation results revealed that the performance of R134a/R1270/R290 (50/5/45 by mass
percentage) was 2.10% higher than the R22 (Sharmas Vali S. and Ashok Babu T.P. 2017). The
present work focuses on the computation of GWP, ODP, RF number and TEWI values of
various R134a/R1270/R290 blends for which authors are not covered in the above literature.
Hence a significant work is identified to compute the above considered parameters. And also
from the above literature COP (3.608) of R134a/R1270/R290 (50/5/45 by mass percentage)
was higher than the COP of R22 (3.534).
2 COMPUTATION OF GWP AND ODP OF R134A/R1270/R290 BLENDS
GWP and ODP values of pure refrigerants (R22, R134a, R1270 and R290) are required to
compute the GWP and ODP of various refrigerant blends. The values of GWP and ODP
are taken from the literature and they are listed in Table 1 (ASHRAE 2009, IPCC 2014). In
the present study total thirteen ternary refrigerant blends (R134a/R1270/R290) at various
compositions are considered and their corresponding designation followed for the blends are
given in Table 1. The correlations used to compute the GWP and ODP of various refrigerant
blends are taken from literature and they are given below (Ahamed J.U. 2014, Arora R.C.
2010).
GWPmix = m1GWP1 + m2GWP2 + … = ∑miGWPi (1)

ODPmix = m1ODP1 + m2ODP2 + … = ∑ m ODP
i i (2)
The GWP and ODP of thirteen investigated blends are computed by using equations (1)
and (2) respectively. The values of GWP and ODP of thirteen studied fluids are also given
in Table 1.
Table 1. Environmental properties of refrigerants.
Refrigerant designation Composition by mass % GWP (100 years) ODP
R22 Pure fluid 1760 0.055

R134a Pure fluid 1300 0
R290 Pure fluid 3 0
R1270 Pure fluid 3 0
M1 (R134a/R1270/R290) 80/10/10 1040.6 0
M2 (R134a/R1270/R290) 60/20/20 781.2 0
M3 (R134a/R1270/R290) 65/5/30 846.05 0
M4 (R134a/R1270/R290) 60/5/35 781.2 0
M5 (R134a/R1270/R290) 55/5/40 716.35 0
M6 (R134a/R1270/R290) 50/5/45 651.5 0
M7 (R134a/R1270/R290) 65/30/5 846.05 0
M8 (R134a/R1270/R290) 60/35/5 781.2 0
M9 (R134a/R1270/R290) 55/40/5 716.35 0
M10(R134a/R1270/R290) 50/45/5 651.5 0
M11(R134a/R1270/R290) 50/15/35 651.5 0
M12(R134a/R1270/R290) 50/20/30 651.5 0
M13(R134a/R1270/R290) 50/25/25 651.5 0
444

3 COMPUTATION OF REFRIGERANT FLAMMABILITY NUMBER
Study of flammability is most crucial for the investigators while developing alternative refrig-
erants from the view point of safety. Flammability gives the range of fuel concentration
within which the refrigerant blends can burn or ignite. These limits are important while com-
puting the hazards of liquid or gaseous fuel mixtures. Hence Jones proposed the correla-
tion to estimate the upper and lower flammability of gases and vapors (Jones, G.W. 1938,
Zabetakis, M.G. 1965). An index named refrigerant flammability (RF) number is used for
indicating the hazards of combustion of the refrigerants. It is reliable to express the haz-
ards of combustion with respect to limits of flammability of each refrigerant by using RF
number. An empirical correlation used for computing the RF number is given below (Shigeo
Kondo et al. 2002).
 U  1/ 2  HOC
RF =   − 1 × (3)
 L   MW
To compute the limits of upper and lower flammability of refrigerant blends, Le Chat-
erier’s rule can be used (Shigeo Kondo et al. 2002).
1 C C C
= 1 + 2 +…= ∑ i (4)
U mix U1 U 2 Ui
1 C C C
= 1 + 2 +…= ∑ i (5)
Lmix L1 L2 Li
From the above literature RF number of R290 and R1270 are 52.2 and 62.1 kJ/g respec-
tively. Flammability limits of R22 and R134a are not available in literature and hence Jones
correlations can be used to compute the flammability limits of R22 and R134a. However
from ASHRAE design safety standard 34, the refrigerants R22 and R134a are classified as
nonflammable ASHRAE A1 category (ASHRAE 34 2007). Based on RF number, refriger-
ants are classified into various groups. If RF number is below 30 then it is considered as
slightly flammable (ASHRAE A2) group and in between 30 to 150 classified as flammable
(ASHRAE A3) group. To compute the RF number of thirteen R134a/R1270/R290 blends
equations from (3 to 5) is used and corresponding values are shown in Table 2.
Table 2. RF number of various studied refrigerant blends
Refrigerants RF number (kJ/g) ASHRAE flammability group
M1 14.90 A2*
M2 20.52 A2*
M3 18.35 A2*
M4 19.79 A2*
M5 21.35 A2*
M6 23.05 A2*
M7 19.57 A2*
M8 21.29 A2*
M9 23.14 A2*
M10 25.15 A2*
M11 23.55 A2*
M12 23.81 A2*
M13 24.07 A2*
*Computed values.
445

4 TEWI ANALYSIS
To compute TEWI of various R134a/R1270/R290 blends the compressor power is required

and it is taken from the literature (Sharmas Vali S. and Ashok Babu T.P. 2017).
Totale quivalent warming index is a measure of impact of global warming based on emis-
sion of greenhouse gases during the working of the refrigeration equipment and the dumping
of the operating refrigerants at the end of life (AIRAH 2012). TEWI takes into an account
of both direct CO2 emissions due to direct emission of refrigerant when the device is oper-
ating with particular refrigerant and indirect CO2 emissions from the various fuels used to
generate electrical energy to operate the apparatus during its lifetime (Thomas W.D. 2004).
TEWI = Direct CO2 emissions + Indirect CO2 emissions

( ) (
= GWP100 × m × L × SLife + E An × SLife × C

) (6)
were GWP100 = GWP of a given fluid for a time period of 100 years, m = Charge of the given
fluid (kg), L = Leakage rate of the refrigerant (%), SLife = Service lifetime of the device (years),
EAn = Energy consumption per annual (kWh), C = Indirect emission factor (kg CO2/kWh).
Table 3. Compressor power of investigated blends.

m Wc WCP
Refrigerants (kg/min) (kJ/kg) (kW)
R22 1.226 38.949 0.7958

M1 1.028 47.366 0.8115
M2 0.894 54.919 0.8182
M3 0.907 52.487 0.7934
M4 0.875 54.032 0.7879
M5 0.847 55.432 0.7825
M6 0.819 57.072 0.7790
M7 0.945 53.754 0.8466
M8 0.919 54.795 0.8392
M9 0.895 56.47 0.8423
M10 0.871 58.215 0.8450
M11 0.829 57.524 0.7947
M12 0.835 57.675 0.8026
M13 0.841 57.828 0.8105
Table 4. TEWI of investigated blends.
Refrigerant TEWI (kg CO2/kWh)
R22 21586
M1 21164
M2 21037
M3 20499
M4 20290
M5 20084
M6 19925
M7 21809
M8 21554
M9 21558
M10 21551
M11 20312
M12 20506
M13 20701
446

The assumptions made while computing the TEWI of various studied refrigerants are
taken from the literature (Vincenzo L.R. 2011). The charge of refrigerant considered in the
device is = 1.5 kg, dischargelevel of refrigerant per annum = 5%, service life of the device = 15
years, operating period of air conditioner = 10 h/d and indirect emission factor = 0.45 kg
CO2/kWh. TEWI values of various refrigerant blends are tabulated in the Table 4.
5.1 GWP of various refrigerant blends

Figure 1 shows the global warming potential of various refrigerant blends. From the fig 1 it
is observed that the GWP of four fluids M6, M11, M12, and M13 are lower among the R22
and thirteen studied fluids. Since all the four refrigerant blends (M6, M11, M12, and M13)
are blended with higher composition of very low GWP refrigerants like R290 and R1270.
5.2 ODP of various refrigerant blends

From the Table 1 it is observed that all the thirteen investigated refrigerants are ozone friendly
in nature. This is due to zero ozone depletion potential of refrigerants (R134a, R1270, and
R290). Since R134a, R1270 and R290 are free from ozone layer depleting substance called
chlorine.
5.3 RF number of various refrigerant blends

Figure 2 shows the refrigerant flammability number of various refrigerants considered in the
present study. Referring to Figure 2 it is evident that RF number of all the thirteen studied
refrigerants are less than 30 kJ/g. Refrigerants with RF number less than 30 are categorized
into mildly flammable (ASHRAE A2). Therefore all the thirteen investigated refrigerant
blends are mildly flammable.
5.4 TEWI of various refrigerant blends

Figure 3 shows the TEWI of various refrigerants investigated in the present study. Refrig-
erants with a low value of TEWI are desirable from the view point of greenhouse effect.
Figure 1. GWP of various refrigerant blends.
447

Figure 2. RF number of various refrigerant blends.
Figure 3. TEWI of various refrigerant blends.
Referring to Figure 3 it is observed that TEWI of M6 is lower among the R22 and thirteen
studied refrigerant blends. This is due to lower compressor power and lower global warming
potential of M6.
6 CONCLUSIONS
From the present study the following can be drawn.

• The COP of M6 was 2.10% higher among the R22 and thirteen studied refrigerants.
• The GWP of M6, M10, M11, M12 and M13 refrigerant blends were lower among the R22
and thirteen studied refrigerants.
• The ODP of all the thirteen studied fluids were zero compared to ODP of R22 (0.055).
• RF analysis exhibited that all the thirteen investigated refrigerant blends were classified
into weakly flammable (ASHRAE A2) group. Since RF number of all the thirteen fluids
were less than 30.
448

• TEWI of ternary blend M6 was lower among the R22 and thirteen studied blends.
• Over all the performance of refrigerant blend M6 (R134a/R1270/R290 50/5/45 by mass
percentage) is better than R22 from the stand point of COP, GWP, ODP and TEWI among
the thirteen investigated fluids and hence it could be an appropriate alternative candidate
to Replace R22 used in air conditioners.
REFERENCES
Ahamed, J.U., Saidur. R, & Masjuki M.M. 2014. Investigation of Environmental and Heat Transfer
Analysis of Air Conditioner Using Hydrocarbon Mixture Compared to R22. Arabian Journal for
Science and Engineering, 39: 4141–4150.
AIRAH. 2012. Methods of calculating Total Equivalent Warming Impact Best Practice Guide lines:
2–20.
ANSI/ASHRAE 2007. Standard 34. Designation and Safety Classification of Refrigerants.
Arora R.C. 2010. Refrigeration and Air conditioning. New Delhi: PHI learning Private Limited.
ASHRAE. 2009. Handbook Fundamentals (SI) chapter 29. Refrigerants: 29.1–29.10.
Atilla Gencer Devecioğlu & Vedat Oruça. 2015. Characteristics of Some New Generation Refrigerants
with Low GWP. Journal of Energy Procedia, 75: 1452−1457.
Chen, W. 2008. A comparative study on the performance and environmental characteristics of R410 A
and R22 residential air conditioners. Applied Thermal Engineering, 28: 1–7.
Devotta, S., Waghmare, A.V., Sawant, N.N., & Domkundwar, B.M. 2001. Alternatives to HCFC-22 for
air conditioners. Applied Thermal Engineering, 21: 703–715.
Global Environmental Change Report, 1997. A brief analysis of the Kyoto protocol. 9(24).
IPCC. 2014. Fifth Assessment Report chapter 8. Anthropogenic and Natural Radiative Forcing:
659–740.
Jones, G.W. 1938. Inflammation Limits and Their Practical Application in Hazardous Industrial Opera-
tions. Chemical. Reviews, 22, 1–26.
Lorentzen, G. 1995. The use of natural refrigerants: a complete solution to the CFC/HCFC predica-
ment. International Journal of Refrigeration, 18 (3): 190–197.
Mohanraj, M., Jayaraj, S. & Muraleedharan, C. 2009. Environment friendly alternatives to halogenated
refrigerants-A review. International Journal of Greenhouse Gas Control, 3(1): 108–119.
Sharmas Vali, S. & Ashok Babu T.P. 2017. Theoretical Performance Investigation of Vapour Compres-
sion Refrigeration System Using HFC and HC Refrigerant Mixtures as Alternatives to Replace R22.
Journal of Energy Procedia, 109: 235–242.
Shigeo Kondo, Akifumi Takahashi, Kazuaki Tokuhashi, Akira Sekiya. 2002. RF number as a new index
for assessing combustion hazard of flammable gases. Journal of Hazardous Materials, A93: 259–267.
Thomas W.D. & Ottone C. 2004. A low carbon, low TEWI refrigeration system design. Applied Ther-
mal Engineering 24: 1119–1128.
United Nations Environmental Programme, (1987). Montreal Protocol on substances that deplete the
ozone layer, Final act. New York: United Nations.
Vincenzo L.R. & Giuseppe P. 2011. Experimental performance evaluation of a vapour compression
refrigerating plant when replacing R22 with alternative refrigerants. Applied Energy 88: 2809–2815.
Zabetakis, M.G. 1965. Flammability Characteristics of Combustible Gases and Vapors. Bulletin 627,
US Bureau of Mines.
449

Indoor performance evaluation of a Photovoltaic Thermal (PVT)

hybrid collector
K.K. Janishali & C. Sajith Babu

Government Engineering College, Calicut, Kerala, India
Sajith Gopi
Kerala Water Authority, Thrissur, Kerala, India
ABSTRACT: A PVT system combines a solar cell, which produces electricity with a solar
thermal collector, which extracts thermal energy from the sun so that it can extract both ener-
gies simultaneously. A solar panel is capable of achieving a maximum electrical efficiency of 15
to 25%, while a PVT hybrid collector produces combined energy efficiency in the range of 55 to
70%. This ability of the PVT collector to trap a large amount of the sun’s energy makes them
superior over the conventional solar panels. This study is aimed to analyse the performance of
a PVT collector under a solar simulator. A 100 watt solar panel is used for the construction of
the PVT collector. Thermal energy is extracted by water flowing through the system, which is in
direct contact with the rear side of the PV panel. Water enters through the inlet pipe, heats up
by absorbing thermal energy from the panel and leaves through the exit pipe. The experiment
was conducted for three light intensities of 600, 800 and 1000 W/m2. The performance analysis
of the PVT system shown that, these systems are four to five times more efficient than normal
PV systems working at outdoor conditions. The results also shown that, the increase in PV
panel temperature results a reduction in electrical efficiency, while the overall efficiency of PVT
system remained almost maximum at all temperatures for a particular light intensity.
1 INTRODUCTION
In this mechanized world, life will be very easy if sufficient amount of energy is available
for us. Every automated system exploits energy to give the desired work output. Therefore,
if energy is not available, every machine is a waste. The overexploitation of non-renewable
energy sources such as fossil fuels will lead to energy crisis. It’s our duty to conserve the avail-
able energy for our upcoming generations by depending on renewable energy sources and
also by utilizing them in an efficient way. While considering the renewable energy sources,
solar energy is considered to be the best among them. This is because, it is sufficiently avail-
able and its tapping is cheaper and easy compared to other energy forms. A photovoltaic
thermal (PVT) system is used to extract energy from sun to a maximum extend. The PV
module will generate electrical energy and the heat developed on the PV module is absorbed
by water or air flowing through the system.
APVT system will bring the advantage of both solar thermal collector and PV panel to
extract thermal as well as electrical energy. This make PVT panels higher efficient when com-
pared to a solar thermal collector or a PV panel individually. A PVT system is an integration
of a PV panel module with an air or water heating system. It produces electrical energy along
with thermal energy. A PVT system consists of a PV panel below which a heat transfer fluid
(air or water) will be flowing to extract the thermal energy from the panel. Since the heat
generated on the PV module is taken away by the heat transfer fluid (HTF), the operating
temperature of the panel will be kept lower than normal conditions so that the panel effi-
ciency is improved.
451

Indoor simulation is used for testing air based PVT collectors (Solanki et al. 2009). From
the test results, an overall energy efficiency of 50% was obtained, out of which electrical effi-
ciency obtained was 8.4% and thermal efficiency was 42%. Thermal efficiency obtained was
considerably higher compared to the electrical efficiency. Sixteen halogen lampseach having a
rating of 500 was used for indoor solar simulation. A small 12v DC fan is used for the circula-
tion of air through the PVT/air system.
A study was conducted (Bahaidarah et al. 2013) to evaluate the performance of PV module
with back surface water cooling. Both numerical and experimental studies of the water based
PVT system were conducted for the hot climate of Saudi Arabia. The working temperature
of the photovoltaic module was significantly reduced about 20%. At the same time the elec-
trical efficiency of the PV module also shown a hike of 9% due to the water cooling provided.
This study was conducted at a light intensity of 900 W/m2 in which about 80% of total energy
was captured by the PVT/water system whereas the solar cell was able to extract only 20% of
the total irradiance.
In another work conducted on a water based PVT system in Saudi Arabia (Harbi, Eugenio
& Zahrani 1998), a higher thermal efficiency is obtained during the summer climate con-
ditions. However the higher temperature of PV panel leads to a loss of 30% in the avail-
able electrical energy. But the overall efficiency of the system remained good due to higher
thermal efficiency. In winter, the PV power output increased even though the thermal side
performance was reduced compared to summer climate.
Researches in the field of PVT collectors started in early of 1940s. Towards the end of 20th
century, researches on PVT systems kept on increasing and the area of interest changed to
building integrated photovoltaic options. This is because PVT systems are aesthetically better
when compared to a separate solar thermal and PV systems.
PVT systems are different type on the basis of coolants used. They are air, water, refriger-
ant and heat-pipe based systems. Air based PVT systems are the most common type used
and they are commercially applied in several engineering practices. These PVT systems can
achieve a maximum of 8% electrical efficiency and 39% of thermal efficiency. Lower heat
removal rate of air based systems is a major disadvantage compared to other systems. Water
based systems are the second popular PVT collector. These systems can achieve electrical effi-
ciency of about 9.5% and thermal efficiency of 50%. When compared to air based systems,
water based PVT collector will provide increased thermal as well as electrical efficiency.
1.1 Indoor simulation

Indoor simulation of solar energy refers to the process of producing illumination in labora-
tory conditions and it facilitates in getting approximate natural light in a controllable condi-
tion. That is, different light intensities can be obtained easily. This method is used in testing
of devices such as photovoltaic panels and sun screens. Light sources like xenon arc lamp and
tungsten halogen lamps are generally used in indoor solar simulation. The solar simulator for
this work consists of 21 halogen lamps connected in series. Each halogen lamp is having a
rating of 150 W. Different light intensity can be obtained using a voltage regulator connected
to rear side of the halogen lamps system.
2 EXPERIMENTAL SETUP
A photovoltaic thermal system is developed with water as the heat transfer fluid. The per-
formance evaluation of the developed system is conducted with the help of indoor simulation
system.
The experimental setup for the testing of PVT system is shown in Figure 1. It consists of
a PVT system, an indoor solar simulator, voltage, current and temperature measuring setup
and a water circulating system consisting of a pump and storage tank.
The PVT system consists of a PV panel of 100 Wand an attachment to the rear side of the
panel to extract thermal energy from it. Energy extraction is achieved by creating a passage
452

through which water flows and keep a direct contact with the rear side of the panel. The
water enters the system through the inlet pipe and heats up by extracting the heat from the
PV module. This will provide hot water and also help to reduce the module temperature so
that its efficiency is increased. The model of the PVT panel showing plastic frame and mica
sheet designed in Creo parametric 2.0 is shown in Figure 2.
Thermocouple are attached on the back side of PV panel at four different places and
paralleled to get the average value of temperature. The terminal of the PV module is taken
outside by drilling hole to the mica sheet. After fixing Mica sheet and connecting the thermo-
couple wires, the gaps are sealed with silicone sealant. Polystyrene foam (Thermocol) sheet is
used to insulate the back side of PVT system. Then GI sheet is screwed on the frame of the
PV panel to complete the design of PVT system.
Power output from the solar panel is obtained using digital multimeter. A variable resistor
is used to study the IV characteristics and obtain maximum power point. A mini-pump is
used to force the water to flow through the PVT system which is controlled with the help of
a temperature controller. A temperature controller is used to control the pump by turning it
on or off according to the pre-set value of temperature. There will be terminals to connect
the sensing thermocouple and a relay switch which will control the pump. The temperature at
which relay turns on is called set point and it can be adjusted with the help of button switches
provided on the temperature controller. It is important to measure temperature at various
areas to study the performance of the developed system. Temperature of the PV module, inlet
and exit temperature of water, and ambient temperatures are measured using K-type ther-
mocouples. A temperature indicator is used to read temperatures from the thermocouples.
Figure 2. Creo design of PVT.
453

2.1 Test procedure
The PV panel and the PVT system are tested in indoor conditions. For testing the PV system,
the PV panel and halogen lamp system are arranged as shown in Figure 1 and the connec-
tions required for temperature and power measurement are also made. Required light inten-
sity is obtained with the help of voltage regulator provided on the halogen lamp system and
measuring with a lux meter. The initial temperature of the PV panel and power output are
measured and noted. Temperature and power variations are noted in a regular interval of one
minute. The testing of the system is continued up to a maximum panel temperature of 70°C
so that higher temperature may cause damage to the PV panel. The experiment is repeated
for different light intensities.
For testing the PVT system, an additional water circulating system is required. After doing
all arrangements similar to PV testing setup, the pump is made on to fill the PVT system
fully with water. Required light intensity is obtained with the help of voltage regulator pro-
vided on the halogen lamp system and a lux meter. The initial temperature of the PV panel
and power output are measured and noted. Temperature and power variations are noted in
a regular interval of one minute. The experiment is continued up to a panel temperature of
55°C so that the efficiency of the PV panel is not reduced much. The experiment is repeated
for different light intensities.
3 DATA REDUCTION
3.1 PV system
To analyze a PV system, it is very simple compared to a PVT system as it includes only electri-
cal energy output. The power output at the maximum power point is used to find the electrical
efficiency of the PV system. The equation for electrical efficiency of the system is given by,
PV power output Imp × Vmp

ηe = = (1)
Energyin G× A
where, Imp and Vmp are the current and voltage at the maximum power point, G is the inci-
dent solar irradiance normal to surface and A is the collector aperture area.
3.2 PVT system

A PVT system provides electrical energy as well as thermal energy as output. The analysis of
the electrical performance of the PVT system is similar to the PV system. Electrical efficiency
of the PV system is given by equation 1 and thermal efficiency of the PVT system is given by,
(Tout − Tin)
Thermal energyoutput mC
ηt = = (2)
Energyin G×A
.
where, m is the mass flow rate of the water through the PVT collector, Tin and Tout are the
temperatures of water at inlet and outlet of PVT panel. The overall performance of the PVT
system is given by the sum of electrical and thermal efficiency.
ηoverall = ηelectricall + ηthermal (3)
4 RESULTS
The PVT and PV systems were tested using indoor solar simulation and the results obtained
at different illuminations are plotted below.
454

The test results of PV and PVT systems at a light intensity of 600 W/m2 are shown in
Figure 3 and Figure 4 respectively.
Efficiency of PV panel decreases with increase in the module temperature. With an increase
of 34°C in PV panel temperature, the PV efficiency shown a relative decrease of 13.2%.
In the case of PVT system, the decrease in overall efficiency is really small. With an increase
of 20ºC in PV panel temperature, only a reduction of 0.18% in overall efficiency is obtained.
A comparison of PVT and PV efficiencies at different panel temperatures with an illumina-
tion of 600 W/m2 is shown in Table 1. The PV panel efficiency is reducing with increase in
temperature while PVT efficiency variation is very low.
Performance of PV and PVT systems at 800 W/m2 is shown in Figure 5 and Figure 6. The
PV and PVT performance at 800 W/m2 is shown a similar trend to that at 600 W/ m2.
The relative PV efficiency is reduced by 17.8% for an increase of 34ºC in PV temperature.
But in the case of PVT system, only a reduction of 1.14% is noted for an increase of 20ºC in
PV temperature.
Table 2 gives the comparison of PV and PVT at different panel temperatures. At 800 W/m2
also PV efficiency is reduced largely compared to the PVT.
Figure 3. PV performance at 600 W/m2.
Figure 4. PVT performance at 600 W/m2.
Table 1. Comparison of PV an PVT at 600 W/m2.
Temperature PV Efficiency PVT Efficiency
35 6.78 65.51
40 6.75 65.50
45 6.67 65.49
50 6.51 65.43
55 6.36 65.39
60 6.22 –
65 6.05 –
70 5.87 –
455

35 6.86 61.39
40 6.55 61.08
45 6.38 60.92
50 6.24 60.77
55 6.16 60.69
60 5.99 –
65 5.82 –
70 5.64 –
The performance of PV and PVT system at 1000 W/m2 is shown in Figure 7 and Figure 8
respectively.
It is clear from figures that the performance of both systems shown a similar trend as of
800 W/m2 and 600 W/m2. There was a reduction of 16.95% in relative performance of the PV
system for a temperature rise of 34°C. In the case of PVT system, the reduction is only 1.17%
of available energy output for a temperature rise of 19°C.
PV and PVT efficiency at different panel temperature is tabulated below for 1000
W/m2 illumination. In this case also, the PV efficiency is significantly affected by temperature
whereas PVT efficiency is not affected much.
From the performance studies of PV and PVT systems at three different light intensities,
it is clear that the PV efficiency increases with increase in light intensity. But the temperature
increase of PV module resulted in reduction of efficiency for all the three light intensities.
In the case of PVT system, the overall efficiency is showing slight decrease with increase
in light intensity. It is because, the corresponding rate of increase in thermal energy is less
compared to rate of increase in light energy obtained with the halogen lamp simulator, when
luminance is increased from 600 W/m2 to 800 W/m2 and then to 1000 W/m2.
Even the PV panel temperature reached up to 55ºC, the temperature attained for water
is only 50ºC to 51ºC only. This is due to the heat transfer loss from PV panel to water.
456

35 6.92 59.77
40 6.82 59.63
45 6.66 59.48
50 6.54 59.30
55 6.30 59.07
60 6.09 –
65 5.91 –
70 5.75 –
Table 4. Flow rate and Tout at different irradiation.
Irradiation intensity W/m2 Flowrate Kg/minute Outlet temperature °C
600 W/m2 0.2143 51
800 W/m2 0.2609 50
1000 W/m2 0.3158 50
The Table 4 will show the flow rates and outlet temperatures of water obtained at three dif-
ferent light intensities for indoor testing.
5 CONCLUSIONS
Performance of the developed PVT system was studied in indoor condition for light inten-
sities of 600, 800 and 1000 W/m2. The variation of electrical efficiency of PV system and
overall efficiency of PVT system were analysed at all three intensities.
457

From the analysis, the overall efficiency obtained for PVT system is very high compared to
the PV panel electrical efficiency. With increase in the temperature, PVT system performed
well whereas the normal PV system suffers to maintain efficiency. That is, PVT overall effi-
ciency maintained almost constant at all temperatures whereas PV efficiency reduced at
higher temperatures. While considering the PVT system, electrical efficiency of PV module
reduced slightly due to increase in module temperature. But a small compromisation in the
electrical efficiency is very well compensated with higher overall efficiency of a PVT system.
The cooling in PVT system will help to maintain the PV panel temperature at a lower value
which will prevent reduction in electrical efficiency.
PV panels normally give an electrical efficiency of 12 to 15%. When compared to that, a
PVT system is 4 to5 times more efficient. It is utilizing less space and is cost effective as com-
pared to separate water heating and PV system. The PVT systems are more visually appealing
and can be integrated to buildings.
NOMENCLATURE
A Collector aperture area m2
C Specific heat of water J/Kg-K
G Incident sola rirradiance W/m2
Imp Maximum powerpoint current A
Isc
. Short circuit current A
m Mass flow rate of water Kg/s
Pmax Maximum Power W
Tin Inlet temperature of water °C
Tout Outlet temperature of water °C
Vmp Maximum power point voltage V
Voc Open circuit voltage V
ηe Electrical efficiency %
ηpower Electrical power generation efficiency %
ηsaving Energy saving efficiency %
ηt Thermal efficiency %
REFERENCES
Adnan Ibrahim & Goh Li Jin. 2009. Hybrid Photovoltaic Thermal (PV/T) Air and Water Based Solar
Collectors Suitable for Building Integrated Applications. American Journal of Environmental Sci-
ences. 5: 614–624.
Al Harbi, Y., Eugenio, N.N. & Al Zahrani, S. 1998. Photovoltaic-thermal solar energy experiment in
Saudi Arabia. Renewable Energy. 15: 483–486.
Bahaidarah, H., Abdul Subhan., Gandhidasan., & Rehman, S. 2013. Performance evaluation of a PV (pho-
tovoltaic) module by back surface water cooling for hot climatic conditions. Energy. 59: 445–453.
Chegaar, M., Hamzaoui, A., Namoda, A., Petit, P., Aillerie, M., & Herguth, A. 2013. Effect of illumi-
nation intensity on solar cells parameters. Advancements in Renewable Energy and Clean Environ-
ment, Energy Procedia. 36: 722–729.
Chow, T.T. 2010. A review on photovoltaic/thermal hybrid solar technology. Applied Energy. 87: 365–379.
Garg, H. & Agarwal, R. 1995. Some aspects of a PV/T collector/forced circulation at plate solar water
heater with solar cells. EnergyConverse Management. 36: 87–99.
Garg, H.P. & Adhikari, R.S. 1997. Conventional hybrid photovoltaic/thermal (PV/T) air heating collec-
tors: steady state simulation, Renewable Energy. 11: 363–85.
Huang, B., Lin, T., Hung, W. & Sun, F. 2001. Performance evaluation of solar photovoltaic/thermal
systems. Solar Energy. 70: 443–448.
Kumar, K., Sharma, S.D. & Jain, L. 2007. Standalone Photovoltaic (PV) Module Outdoor Testing
Facility for UAE Climate. CSEM-UAE Innovation Center LLC 2007.
Solanki, S.C., Swapnil Dubey & Arvind Tiwari. 2009. Indoor simulation and testing of photovoltaic
thermal (PV/T) air collectors. Centre for Energy Studies, Applied Energy, IIT Delhi. 86: 2421–2428.
Xingxing Zhanga., Xudong Zhaoa., Stefan Smitha., Jihuan Xub. & XiaotongYuc. 2012. Review of
R&D progress and practical application of the solar photovoltaic/thermal (PV/T) technologies.
Renewable and Sustainable Energy Reviews. 16: 599–617.
458

Properties of biodiesel and blends: An investigative and

comparative study
P.P. Yoosaf, C. Gopu & C.P. Sunil Kumar

ABSTRACT: Biodiesel is a renewable energy source and an alternate fuel for compres-
sion ignition engines as an alternative for diesel. Biofuel satisfies the physical and chemical
standards of diesel. Hence it can be used as an alternative for diesel in compression ignition
engines. Compared with normal diesel, biodiesels cause less pollution. The main objective
of this experimental study is to compare the various properties of pure diesel, fish biodie-
sel, coconut testa biodiesel, coconut testa Biodiesel-Ethanol-Diesel blend (BED), which are
made up of 5 vol% of ethanol and 10 vol% of biodiesel to 85 vol% diesel fuels, and coconut
testa Biodiesel-Ethanol blend (BE), which are made up of 5% ethanol and 95% biodiesel.
Properties such as flash-fire point, density, viscosity, Acid Value (AV), Saponification Value
(SV), Iodine Value (IV), and calorific value of different biodiesels were analyzed.
Keywords: Biodiesel, Transesterification and Properties
1 INTRODUCTION
Recent development in the applications of alternative fuels for compression ignition engines have
attracted attention in the automobile domain due to the depletion of fossil fuels and increasing
air pollution problems caused by the emissions from various engines. The plastic pyrolysis oil
is another alternative fuel for some engine application in certain operation conditions. Biodiesel
consists of a mixture of ethyl or methyl esters of fatty acids derived from vegetable oils or animal
fats, which are obtained from the transesterification reaction with short-chain alcohol, methanol
or ethanol, respectively and in the presence of a catalyst (Parente, 2003). The properties, namely
density, viscosity, flash point and fire point of fish oil biodiesel are higher and the calorific value
is 0.92 times that of diesel (Shivraj et al., 2014). Flash and fire points increase with an increase in
the amount of biodiesel in the blend. The cetane number of Fish Oil Biodiesel (FOB) is higher;
this ensures the complete combustion of FOB. The calorific value of B100 is less and increases
with the increase in the amount of diesel fuel in the blend, and the flash and fire points also
increase with increase in the amount of biodiesel in the blend (Pavan & Venkanna, 2014).
Fuel stability related properties, acid value and iodine value of testa biodiesel is within the
range, which shows it has good storage stability (Swaroop et al., 2016). The best proposed
solution for reducing diesel engine pollutants is using biofuels that consist of a combina-
tion of diesel, biodiesel and ethanol (Hoseini, 2017). A mixture of biodiesel-diesel-ethanol
blend is utilized to increase the poor cold-flow properties of biodiesel as the cetane number
and lubricity of ethanol-diesel blends is too low (Hatkard et al., 2015). The blended fuels
reduced PM emissions, while increased NO x emissions, but reduce smoke and CO emissions
(Çelikten, 2011). Using ethanol as fuel or a fuel additive in diesel engines is limited by their
miscibility problems with diesel fuel. Other problems are low their cetane number, low lubric-
ity and reduced heating value (Altun et al., 2011). In comparison with the diesel fuel, biodie-
sel blends produced lower sound levels due to many factors, including an increase in oxygen
content, reduction in the ignition delay, higher viscosity and lubricity (Liaquat et al., 2013).
The use of different vegetable oils affects production processes and costs, and the resulting
459

biodiesel characteristics used cooking oil require pre-treatment prior to traditional alkali-
catalyzed transesterification (Titipong & Ajayk, 2014). The thin brown colored layer outside
the fleshy white kernel of coconut when peeled out and pressed produces the coconut testa
oil. The testa oil thus produced is saturated and non-edible. Coconut testa is a waste product
in the coconut processing industries. It is after peeling out the testa the industries producing
products like coconut milk, virgin oil etc. It is also used as a cattle feed. Oil produced from
testa is used in paints in industries. This work also aims at producing biodiesel from coconut
testa. Due to the wide availability and non-edible nature of the oil, biodiesel produced from
testa has enormous potential in meeting the future fuel demands. Being saturated and non-
edible, it is one of the best feedstocks available for biodiesel production.
2 MATERIAL AND METHODS
2.1 Production of biodiesel

Arbee Biomarine is the largest producer and manufacturer of biodiesel and glycerin in Ker-
ala, India. The company’s biodiesel refinery employs the latest refining technology and equip-
ment and quality assurance procedures. Arbee biodiesel can be directly used in pure form,
or in combination with petrodiesel. With virtually no sulphur content, biodiesel has better
lubricating properties and much higher cetane ratings. Biodiesel is also proven to reduce fuel
system wear. Arbee Biomarine’s biodiesel are ideally suited for diesel engines. Biodiesels have
much more applications in boilers.
2.2 Transesterification reaction

Transesterification is the process of the reaction of triglyceride (oil/fats) with an alcohol in
the presence of an acidic or alkaline catalyst, and it requires the reaction temperature to be
below the boiling point of the alcohol used The reaction time should not be less than 30 min-
utes or more than two hours to form mono alkyl ester that is biodiesel and glycerol. It is a
widely employed procedure to reduce the high viscosity of triglycerides.
Triglycerides + Methanol → Methyl ester + Glycerol (Catalyst – KOH).
This is the method used for the preparation of biodiesel from testa. 1 liter of coconut testa oil
was taken in a conical flask. 250 mL methanol taken in a beaker and 8 g of KOH was added to it.
KOH was allowed to dissolve in methanol to form methoxide. The coconut testa oil was heated
to 60°C with constant stirring. The methoxide solution was poured into the oil. The solution was
stirred continuously for 45 minutes. The resulting mixture was poured into a separating flask and
kept for 24 hours. After 24 hours the glycerol settled at the bottom of the flask with biodiesel on
the top. The glycerol settled at the bottom of the flask was separated out to obtain the biodiesel.
3 EXPERIMENTAL PROCEDURES
3.1 Density
Density was measured using the standard method (BIS, 1972). A capillary stopper relative
density bottle of 50 ml capacity was used to determine the density of the biodiesel. Density
was calculated using the following equation.
W3 − W1
Density = × ρ H 2O (1)
W2 − W1
3.2 Kinematic viscosity

Viscosity is a measure of resistance to flow of a liquid due to internal friction caused by one
part of a fluid moving over another. Viscosity of biodiesel is measured using a Brookfield
460

viscometer, which satisfies the ASTM standard. The Brookfield viscometer consists of a cylin-
drical container with a metallic spindle rotating within. For the measurement of viscosity, the
biodiesel is poured into the container up to the mark. The spindle rotates within the cylinder
at a constant rpm. The instrument measures the torque and thus calculates the viscosity.
3.3 Iodine value

Iodine value or iodine number was introduced in biodiesel quality standards for evaluating its
stability to oxidation, and is expressed as the g of iodine that will react with 100 g of biodie-
sel. A higher iodine value indicates more unsaturated content. Iodine value is a measurement
of total unsaturation of fatty acids. Biodiesel having high iodine value is easily oxidized in
contact with air. The iodine value largely depends on the nature and ester composition of the
feedstocks used in biodiesel production.
B = Volume of sodium thiosulphate for blank
S = Volume of sodium thiosulphate for sample
W = Weight of sample, g
Iodinevalue =
( B − S ) × N ×12.69 (2)
W
3.4 Acid value

The acid number or neutralization number is a measure of the amount of free fatty acids
contained in a fresh fuel sample and of free fatty acids and acids from degradation in aged
samples. Acid value determination is used to quantify the presence of acid moieties in a
biodiesel sample. The acid number determines the degree of degradation of biodiesel when
the fuel is used. The acid value of biodiesel is defined as the number of milligrams of NAOH
required to neutralize free acid present in one gram of biodiesel.
N × V × 56.1
Acid value = (3)
W
N = Normality = 0.1
V = Volume of NAOH required in mL
W = Weight of sample in g
3.5 Saponification value

The process formation of soap is called saponification. The saponification value is defined as
the milligrams of KOH required to saponify 1 g of fat or oil.
N × ( b − s ) × 56.1
Saponification Value = (4)
W
W = Weight of the sample

B = Titre value for blank
S = Titre value of sample
3.6 Flash point and fire point

The temperature at which the vapor of a liquid flash when subjected to a naked flame is
known as the flash point of the liquid.
The flash point is used to classify fuel for transport, storage and distribution according to
its hazard level.
461

3.7 Calorific value
The calorific value is the amount of heat energy released by the combustion of unit mass of
a fuel. Fuel has higher calorific value and lower calorific value. The amount of heat released
as H2O after the combustion is condensed to liquid state is called higher calorific value. The
amount of heat released as H2O after combustion is in the vapor state is called lower calorific
value. The calorific value of fuel was determined by using IKA C200 bomb calorimeter.
4.1 Density
The density of different fuels is shown in Figure 1 and it is different for each fuel sample. The
value of density of BED blend is 828.37 kg/m3, and the density of testa biodiesel is 832.3 kg/m3,
both of which are comparable with the density of diesel (833 kg/m3). The density of BE blend
is 875 kg/m3 and for fish biodiesel the density is found to be 787.5 kg/m3. All these values
satisfy ASTM standards (575–900 kg/m3).
4.2 Kinematic viscosity

The kinematic viscosity of different fuels are shown in Figure 2 and it is different for each
fuel sample. The value of kinematic viscosity of BED blend is 2 cP, and the density of testa
biodiesel is 5.9 cP. The kinematic viscosity of fish biodiesel is comparable with the kinematic
viscosity of diesel (2.95 cP). All these values satisfy ASTM standards (1.9–6).
Figure 1. Density of different fuels.
Figure 2. Viscosity of different fuels.
462

4.3 Flash point and fire point
The flash point and fire point of different fuels are shown in Figure 3 and it is different for
each fuel sample. The flash point and fire point were determined for all investigated fuels and
it is found that biodiesel blended with ethanol has the lowest flash and fire points. Lower
values indicate that those fuels are difficult to handle and transport. The flash and fire point
of pure biodiesels, such as fish biodiesel and testa biodiesel, are much higher than pure diesel.
4.4 Iodine value

The iodine value of different fuels are shown in Figure 4 and it is different for each fuel sam-
ple. It is found that diesel has the lowest iodine value at 2.66 mg I/g. All biodiesels have higher
iodine values than the pure diesel, which indicates the fuel samples have more unsaturated
contents.
4.5 Acid value

The acid value of different fuels are shown in Figure 5 and it is different for each fuel sample.
The maximum value of acid value for biodiesels based on ASTM is 0.5 mgKOH/g. All the
investigated fuels meet the standard value. BED and BE blend have almost equal and higher
values. The acid value is not applicable for pure diesel since it has a lower degree of degrada-
tion due to the low amount of carboxylic acid groups.
Figure 3. Flash and fire point of different fuels.
Figure 4. Iodine value of different fuels.
463

4.6 Saponification value
Saponification of different fuels are shown in Figure 6 and it is different for each fuel sample.
Saponification is minimal for pure diesel and much higher for all other biodiesel samples.
The maximum value is found in fish biodiesel of the samples. All these values satisfy ASTM
standards (max 120).
4.7 Calorific value

The calorific value of different fuels are shown in Figure 7 and it is almost equal for each fuel
sample. Diesel fuel has the maximum value and all other biodiesel samples have a low and
almost equal values.
Figure 5. Acid value of different fuels.
Figure 6. Saponification value of different fuels.
Figure 7. Calorific value of different fuels.
464

5 CONCLUSION
The proper constant quality of biodiesel can only be promising by analyses the biodiesel
quality standards like ASTM, EN and BIS. To attain this aim it is very important to check
the quality throughout the biodiesel production, such as trans-esterification, emulsification
or any other production techniques, and from the feedstock to the distribution units. The
physical and chemical properties of different biodiesels are mainly influenced by the com-
position of feedstock used in their production process, and the nature of the feedstock and
its storage conditions, such as air, sunlight and humidity. Furthermore, different area mar-
kets need different quality requirements. The main differences are found in viscosity, iodine
value, density and acid value. Various other reasons encountered for these variations are
the performance describing properties at very low temperature, such as density at 15°C, as
per ASTM standards, and the exposed conditions. It is not possible to devise a formula for
standards of biodiesels due to these differences. This would be a major disruption for both
biodiesel imports and exports among different countries of the world, and also for the auto-
motive industry.
REFERENCES
Altun, S., C. Öner, F. Yaşar, & Firat, M. (2011). Effect of a mixture of biodiesel-diesel ethanol as fuel on
diesel engine emissions. International Advanced Technologies Symposium (IATS’11), Elazığ, Turkey,
16–18.
Çelikten, I. (2011). The effect of biodiesel, ethanol and diesel fuel blends on the performance and
exhaust emissions in a DI diesel engine. Gazi University Journal of Science, 24(2), 341–346.
Hatkard, N., Salunkeg, B. & Lawande, V.R. (2015). The impact of biodiesel-diesel-ethanol blends.
volume 2, Issue 5.
Hoseini, S. (2017). The effect of combustion management on diesel engine emissions fueled with biodie-
sel-diesel blends; Renewable and Sustainable Energy Reviews, 73, 307–331.
Liaquat, A.M., Masjuki, H.H., Kalam, M.A., Rizwanul Fattah, I.M., Hazrat, M.A., Varman, M.,
Mofijur, M. & Shahabuddin, M. (2013). Effect of coconut biodiesel blended fuels on engine perform-
ance and emission characteristics. Procedia Engineering, 56.
Parente, E.J. (2003). Biodiesel: uma aventura tecnológica num país engraçado (1st ed). Fortaleza:
Unigráfica.
Pavan, P. & Venkanna, B.K. (2014). Production and characterization of biodiesel from mackerel fish oil.
International Journal of Scientific & Engineering Research, 5(11).
Shivraj, H., Astagi, V. & Omprakash, D.H. (2014). Experimental investigation on performance, emission
and combustion characteristics of single cylinder diesel engine running on fish oil biodiesel. Inter-
national Journal for Scientific Research & Development (IJSRD), 2(7), ISSN (online): 2321-0613.
Swaroop, C., Tennison, K. Jose, & Ramesh, A. (2016). Property testing of biodiesel derived from coco-
nut testa oil and its property comparison with standard values, ISSN, 2394–6210, 2(2).
Titipong, I. & Ajayk, D. (2014). Biodiesel from vegetable oils. Renewable and Sustainable Energy
Reviews, 31, 446–471.
465

Thermal sciences and transport phenomena

Computational study on two phase flow of liquid nitrogen

with different internal coatings
S.S. Bindu, Sulav Kafle, Godwin J. Philip & K.E. Reby Roy
TKM College of Engineering, Kollam, Kerala, India
ABSTRACT: One of the most important application of cryogenics includes the transfer of
cryogenic fluid from storage site to its utilization. To optimize the initial phase of cryogenic
heat transfer, twisted channels are coated with different coating materials, which increases the
chill down efficiency of cryogenic systems. Applications of coating materials like graphene,
CNT, polyurethane and Teflon on twisted channels has significant time saving in cool down
compared to conventional channels surfaces. Computational study was performed to evaluate
the enhancement of heat transfer and coating effectiveness on chill down time. The chill down
of two surfaces uncoated and coated is compared and latter one is found to be more efficient.
Keywords: Chill down, Cryogenics, Liquid Nitrogen, Nucleate boiling, Polyurethane coat-
ings, Twisted channels
NOMENCLATURE
Ta External temperature
Ti Coil initial temperature
T Inlet temperature
qc, qu heat load on coated and uncoated panels.
Uc, Uu overall heat transfer coefficient for coated and uncoated panels.
(∆θ)c, (∆θ)u temperature difference between skin temperature and LN2 for
coated and uncoated panels.
Ac, Au panel area for the coated and uncoated panels.
1 INTRODUCTION
Cryogenic liquids are used in many technological applications: such as propulsion systems,
cooling of superconducting magnets etc. It is also being widely adopted for various clinical
applications. Cryogen transfer involving two-phase flow is an indispensable procedure before
the operation of these systems. This transfer process is characterized by their highly transient
nature. Cryogenic chill down refers to the process by which the temperature of the transfer
line is lowered to the saturation temperature of cryogen. This process is highly unstable and
characterized by large pressure fluctuations accompanied by transient boiling heat transfer.
A team of researchers at MIT (Preston, Mafra, Miljkovic, Kong, & Wang, 2015) studied
the usefulness of ultrathin grapheme coatings on conducting materials using CVD process
and reported that it promotes drop wise condensation. CFD analysis of single-phase flows
through coiled tubes was (Jayakumarar et al. 2010) and found that the fluid particles undergo
oscillatory motion inside the pipe causing fluctuation in heat transfer rates.
Numerical investigation on heat transfer from hot water in shell to cold water flowing in a
helical (Neshat, HossainpourF, & Bahiraee, 2014) coil was made and identified that the mass
flow rate and specific heat of fluids are dependent on the shell side fluid temperature and
geometric parameters of coil.
469

Goli et al. (2013) conducted an experiment and demonstrated that chemical vapor deposi-
tion of graphene on Copper films strongly enhance their thermal diffusivity and thermal con-
ductivity. The enhancement is primarily due to changes in Cu morphology during graphene
deposition and associated with temperature treatment. Allen et al. conducted an experiment
in which (Allen, 1965) they observed a sudden drop in test section wall temperature at the
start of the chilldown process. They concluded it as the result of higher rate of cryogenic
heat transfer between transfer line wall and the cryogen being transported. After a period of
time, the rate of decrease in wall temperature remains approximately continual until the entire
test line is cooled down. In coated panels, the initial heat transfer rate is reduced due to the
thermal resistance offered by the Teflon coating and the gas film to the cryogen. As the fluid
volume increases, there arises a thermal gradient on each side of the coating leading to the
faster attainment of nucleate boiling on the surface in contact with the cryogen. This results
in shorter chill down time.
Equations 1 and 2 helps in the comparison of overall heat transfer coefficients under
steady state for coated and uncoated panels.
qc = UcAc ( ∆θ ) c (1)
qu = UuAu ( ∆θ ) u (2)
The coatings on good conductors provide rapid lowering in surface temperature

which, in turn permits rapid attainment of liquid solid contact with good heat removal
rates (Maddox, 1966). Comparison of cool down time between internally coated and
uncoated propellant lines (Reed, Fickett, & T, 1967) was conducted and found that Kel-F
coated pipe cool down rate is faster than that of uncoated line. Sheffer et al. conducted
two phase flow of LN2 with pulsated flows and arrived at a conclusion that (Shaeffer
et al. 2013) flows with higher Reynolds number will results in higher efficiency. Reed
et al. (1967) reported that unlike in uncoated transfer lines where, immediately after cryo-
gen entry into the transfer line a thin layer of vapor is established between the cryogen
and the pipe wall causing an increase in chill down time, a non-conductive coating on a
conducting pipe can reduce chilldown time by creating a thermal gradient between the
fluid-fluid interface.
2 NUMERICAL STUDY USING COMSOL
Heat transfer analysis in twisted channel and effect of coating materials are analyzed
by varying coating material, thickness of coating and fluid flow rates. Coating materials
used are Polyurethane, Teflon, Graphene and CNT. Flow rates are varied from 0.1 m/s to
0.01 m/s. Coating thickness are varied from 0.025 mm to 0.1 mm for Polyurethane and
Teflon, and 0.1 mm for Teflon and CNT. The properties of coating materials are shown
in Table 1.
Table 1. Properties of coating materials.
Thermal conductivity (k) Specific heat (Cp) Density

Material/properties W/m.K J/kg.K kg/m3
Polyurethane 0.026 1.76 30

Teflon 0.25 970 2200
Graphene 2000 730 1500
CNT 3000 450 1300
470

2.1 Heat transfer in twisted coil
Steps involved are as follows:
• First model is drawn by drawing two right handed helices giving axial pitch as 50 mm,
major and minor radius as 100 mm and 10 mm with a single number of turn.
• Coil material used is copper and a thin layer is added to the inner wall of the coil. The
properties of the thin layer are given that of different coating material with all the follow-
ing properties.
• The flow of fluid is taken as Turbulent with time dependent under k-epsilon model.
• Mesh used is physics controlled coarse and mesh statistics are as follows:
The number of Tetrahedral, Prism, Triangular, Quadrilateral, Edge and Vertex elements are
19240, 15250, 5494, 160, 863 and 16 respectively.
• Study is done using surface average to find coil surface temperature.
2.2 Boundary conditions

For all case LN2 is used as a fluid, the other boundary conditions used are as follows:
Inlet temperature is taken as 77 K with air as external convection medium. External and
initial Coil temperature was 300 K. The Coating thickness for Polyurethane and Teflon was
taken as 0.025 mm and 0.1 mm respectively. The corresponding values for Graphene and
CNT was 0.1 mm.
3 ANALYSIS
ase 1: (a) Velocity = 0.01 m/s, Coating of polyurethane having thickness 0.025 mm

3.1 C
and 0.1 mm
Figure 1 shows surface temperature variation along the coil. It depicts that the temperature
goes on increasing as very cold liquid nitrogen is passed through inlet, lower right end.
Temperature increases as there is heat transfer from outside hot region to inner cold fluid.
Figures 2 and 3 depicts the variation of average surface temperature with time when LN2
is flowing through the coil at the flow rate of 0.01 m/s. Average total surface temperature of
coated tube is decreased as cold fluid LN2 is passed through it. In case of coating thickness
0.025 mm, it is seen that the tube surface temperature is 267 K at 56.7 second for uncoated
coil and same temperature is attained at 56.3 second for coated one. Coated one takes 0.7%
lesser time to reach that temperature. For a coating thickness of 0.1 mm, the tube surface
temperature is 268 K at 56.7 seconds for uncoated coil and same temperature is attained at
56 seconds for coated one. Coated one takes 1.25% lesser time to reach the same temperature.
Figure 1. Close view of temperature contour.
471

Figure 2. Variation of surface temperature with time when inlet velocity is 0.01 m/s, coating thickness
0.025 mm.
0.1 mm.
3.2 Case 1:(b) Velocity = 0.01 m/s, Coating of Teflon having thickness 0.025 mm
The result obtained is:
In this case, from Figure 4, it is seen that the tube surface temperature is 272.88 K at 40 sec-
onds for uncoated coil and same temperature is at 39.3 seconds for coated one. So, coated coil
takes 0.7 seconds i.e. 1.75% lesser time to reach that temperature.
Figure 5 depicts the variation of velocity magnitude at various section slices along the coil
when liquid nitrogen is passed through at a flow rate of 0.01 m/s. It shows velocity is goes on
decreasing from inlet to outlet. Also at all sections, velocity is higher in outward direction,
this may be due to centrifugal action of fluid.
Figure 6 shows temperature gradient at different time t = 0, 5 and 10 s. Temperature gradi-
ent is greater along interface surface of solid and fluid as this is the area where heat transfer
interaction occurs.
3.3 Case 1: (c) Velocity = 0.01 m/s, Coatings of Graphene and CNT having thicknesses 0.1 mm
The geometric model for this case is shown in Figure 7. From Figure 8 it is seen that the tube
surface temperature is 277 K at 28 seconds for uncoated coil and same temperature is at 23.1 sec-
onds for graphene coated. Similarly, 277 K is at 20.9 seconds for CNT coated coil. So, graphene
coated coil takes 18.2% lesser time and CNT coated takes 25.35% to reach that temperature.
472

Figure 4. Variation of surface temperature with time.
Figure 5. Contour of velocity magnitude.
Figure 6. Temperature gradient at inlet section when inlet velocityis 0.01 m/s.
Figure 9, depicts temperature gradient variation at inlet section slice of the coil at different
time, t = 1, 10, 20, 30, 40, and 50 s. Temperature gradient is greater along interface surface
of solid and fluid as this is the area where heat transfer interaction occurs. Figure 10 shows
surface temperature variation along the coil. It depicts that the temperature goes on increas-
ing as very cold liquid nitrogen is passed through inlet (lower right end in above figure).
Temperature increases as there is heat transfer from outside hot region to inner cold fluid.
Figures 11 and 12 depicts the variation of average surface temperature with time when LN2
473

Figure 8. Variation of surface temperature with time.
Figure 9. Close view of temperature gradient contour.
474

0.025 mm.
0.1 mm.
is flowing through the coil at the flow rate of 0.1 m/s. Average total surface temperature is
decreased when cold fluid LN2 is passed through the tube.
ase 2: (a)Velocity = 0.1 m/s, Coating of polyurethane having thickness 0.025 mm and

3.4 C
0.1 mm
In case of coating thickness 0.025 mm, the tube surface temperature is 268.1 K at 59.9 sec-
onds for uncoated coil and same temperature is at 56 seconds for coated one. So, coated coil
take 3.9 seconds i.e. 6.5% lesser time to reach that temperature. Again, in case of coating
thickness of 0.1 mm, it is seen that the tube surface temperature is 268.1 K at 55.7 seconds
for uncoated coil and same temperature is at 59.9 seconds for coated one. So, coated coil take
4.3 seconds i.e. 7.17% lesser time to reach that temperature.
Figure 13 depicts the variation of velocity magnitude at various section slices along the
coil when liquid nitrogen is passed at flow rate of 0.1 m/s. It shows velocity goes on decreas-
ing from inlet (lower part in above figure) to outlet. Also at all sections, velocity is higher in
outward direction, may be due to centrifugal action of fluid.
Figure 14 shows temperature gradient at different time t = 1, 5 and 10 s. Temperature gra-
dient is greater along interface surface of solid and fluid as this is the area where heat transfer
interaction occurs.
Figure 15 shows surface temperature variation along the coil. It depicts that the tempera-
ture goes on increasing as very cold liquid nitrogen is passed through inlet (lower right end
in above figure). Temperature increases as there is heat transfer from outside hot region to
475

Figure 14. Temperature gradient at inlet section when inlet velocity is at 0.01 m/s.
inner cold fluid. Figure 16 depicts the variation of average surface temperature with time
when liquid nitrogen is flowing through the coil at the flow rate of 0.1 m/s. Average total
surface temperature is decreased when coated as cold fluid LN2 is passed through the tube.
In case of coating thickness 0.025 mm, it is seen that the tube surface temperature is 268.1 K
at 59.9 seconds for uncoated coil and same temperature is at 56.4 seconds for coated one. So,
coated coil take 3.5 seconds i.e. 5.85% lesser time to reach that temperature. Again, in case of
coating thickness 0.1 mm, it is seen that the tube surface temperature is 268.1 K at 57.6 sec-
onds for uncoated coil and same temperature is at 59.9 seconds for coated one. So, coated coil
take 2.3 seconds i.e. 4% lesser time to reach that temperature.
476

0.025 mm.
3.5 Case 2: (b) Velocity = 0.1 m/s, Coating of Teflon having thickness 0.025 mm and 0.1 mm
Figure 17 depicts the variation of velocity magnitude at various section slices along the coil
when liquid nitrogen is passed at a flow rate of 0.1 m/s. It shows that velocity goes on decreas-
ing from inlet (lower part in above figure) to outlet. Also at all sections, velocity is higher in
outward direction, may be due to centrifugal action of fluid.
4 CONCLUSION AND SCOPE OF FUTURE WORK
From numerical study, by using polyurethane, Teflon, CNT and graphene, the cool down
rate is significantly lesser for coated helical channels than uncoated ones. These results should
prove useful in the design of future transfer lines. Future work can employ different coating
materials to cool down transfer lines made of stainless steel and other suitable materials with
different coil shapes.
477

ACKNOWLEDGEMENT
The authors would like to acknowledge the Space Technology laboratory of TKM college
of Engineering and also the Kerala State Council for Science Technology and Environment
(KSCSTE) for providing facilities for the successful completion of the project.
REFERENCES
Allen, L.D. 1965. Advances in Cryogenic Engineering. Texas: Cryogenic Engineering Conference Rice
University Houston.
Goli, P., Ning, H., Li, X., Lu, C.Y., Novoselov, K.S., & Balandin, A.A. 2013. Strong Enhancement of
Thermal Properties of Copper Films after Chemical Vapor Deposition of Graphene.
Jayakumarar, J.S.S.M. Mahajania, J.C. MandalaKannan, N. Iyera, & P.K. Vijayan. 2010. CFD analysis
of single-phase flows inside helically coiled tubes. Computers & Chemical Engineering, 34:430–446.
Maddox, J.P. 1966. Advances in Cryogenic Engineering. (pp. 536–546). New York: Plenum Press.
Neshat, E., Hossainpour F.S., & Bahiraee. 2014. Experimental and numerical study on unsteady natural
convection heat transfer in helically coiled tube heat exchangers. Heat and Mass Transfer, 50 (6):
877–885.
Preston, D.J., Mafra, D.L., Miljkovic, N., Kong, J., & Wang, E.N. 2015. Scalable Graphene Coatings for
Enhanced Condensation Heat Transfer. American Chemical Society, 15: 2902–2909.
Reed, R.P., Fickett, F., & T, L. 1967. Advances in Cryogenic Engineering. 12, pp. 331–339. Plenum Press.
Shaeffer. R, Hu. H, & Chung. J.N. 2013. An experimental study on liquid nitrogen pipe chilldown and
heat transfer with pulse flows. International Journal of Heat and Mass Transfer, 67: 955–66.
478

An experimental investigation of cryogenic chilldown time

in a polyurethane-coated helical coil
Jesna Mohammed, Pranav Shyam Praveen & K.E. Reby Roy

Thangal Kunju Musaliar (TKM), College of Engineering, Kollam, Kerala, India
ABSTRACT: To optimize the cryogenic chilldown of transfer lines and to improve the effi-
ciency of cryogenic systems, chilldown time should be reduced. Such time saving is associated
with reduced consumption of cryogenic fluids. Experiments were performed on helical channels
with a polyurethane coating for different inlet pressures, and the performance of an untreated
helical surface and one with a coating were compared at corresponding pressures. The results
indicated that there are substantial savings in chilldown times with coated surfaces compared to
non-coated ones. Liquid nitrogen was used as the cryogen and was passed through helical coils
made of copper. The significant reduction in chilldown time is observed only after the onset of
a nucleate boiling regime, as indicated by a graph of average surface temperature versus time.
Keywords: Chilldown, Cryogenics, Helical channels, Liquid nitrogen, Polyurethane coating
1 INTRODUCTION
The scope of cryogenic fluid use is typically in industries, space exploration, cooling of electronic
components, and in the medical field. Transfer of cryogens to their associated installations is
very important prior to their operation. Cryogen transfer is accompanied by phase changes in
flow, pressure surges and flow reversal. When cryogens are introduced into a transfer line that is
in thermal equilibrium with the ambient temperature, uncontrolled evaporation occurs. In order
to establish a steady flow of fluid during this initial phase, cooling down of equipment is a pre-
requisite, which is termed cryogenic chilldown. To design a cryogenic transfer line, phenomena
such as heat transfer, fluid flow, and changes in pressure across the test line need to be identified
and managed. Observing chilldown processes, Yuan et al. (2008) indicated that the cryogenic
liquid encounters three boiling regimes during chilldown: film, transition, and nucleate boiling.
This differs from boiling experiments in that no external heat is provided in the test section.
Berger et al. (1983) found that helically coiled tubes are superior to straight tubes for heat
transfer applications. Because of the curvature of helical coils, centrifugal force is introduced
resulting in the development of secondary flows, as reported by Dravid et al. (1971). Thus,
the movement of the outermost fluid tends to be faster than that at the inside of the coil,
which increases the turbulence and thereby increases heat transfer. As reported by Cowley
et al. (1962), a reduction in the time taken for cooldown of cryogenic equipment was seen
when metallic components were coated with materials of poor thermal conductivity. Allen
(1966) suggested that heat transfer by virtue of forced convection between the entering liq-
uid and the transfer line wall results in a sudden temperature drop during the initial phases
of the chilldown process. After a brief time period, the rate of temperature drop reduces to
a minimum and is then maintained until chilldown is attained. During this period, the flow
encountered is film boiling with relatively low-velocity gas flow. When low-conducting coating
materials such as Teflon are introduced between the transfer line wall and the cryogen, a ther-
mal gradient is developed resulting in the early attainment of a temperature corresponding
to a nucleate boiling regime. This eventually results in higher rates of heat transfer, leading to
faster chilldown of the line. Maddox and Frederking (1966) reported that heat removal rates
479

can be improved by providing a coating on good conductors. Leonard et al. (1967), through a
comparative study of the chill down times of transfer lines with and without coatings under
low-velocity fluid flow, concluded that the coating can enhance heat transfer. In their study
on Critical Heat Flux (CHF) in helically coiled tubes, Jensen and Bergles (1981) concluded
that CHF initially increases with mass velocity, but then decreases after achieving a maximum
value; they found that CHF occurs at the inside surface first. Because of higher centrifugal
force and secondary flow, a coil with smaller radius was found to have a higher critical value
for the same heat flux, mass velocity and tube length.
In a comparison of heat transfer rates between a straight tube and a helically coiled heat
exchanger, Prabhanjan et al. (2002) concluded that helical coils have better heat transfer char-
acteristics and that geometry and flow rate play significant roles in elevating the temperature
of the fluid.
In a study of the thermochemical characteristics of refrigerant R134a flow boiling in heli-
cally coiled tubes at low mass flux and low pressure, Chen et al. (2010) found that the heat
transfer coefficient increases with rises in mass flux, vapor quality, and pressure. Their study
also provided many correlations for helical flow.
By conducting transient cryogenic chilldown processes in horizontal and inclined pipes, John-
son and Shine (2015) proposed an upward inclination for effective cryogen transport. In review-
ing the two-phase heat transfer characteristics of helically coiled tube heat exchangers, Fsadni
and Whitty (2016) suggested some practical correlations for two-phase heat transfer coefficients.
Research by Hardik and Prabhu (2017) on the critical heat flux in helical coils at low pres-
sure done on film boiling crisis concluded that CHF increases as the mass flux increases and
decreases with increase in quality and, for the same quality, CHF may decrease with increase
in tube and coil diameter.
Thus, the present experiment combines two chilldown strategies by creating a helical
geometry and coating it with polyurethane, and the objective of this research is to prove that
coated coils are more efficient for use in cryogenic system installations than uncoated ones.
2 EXPERIMENTAL APPARATUS
2.1 Basic experimental facility

The experimental setup consisted of a gaseous nitrogen cylinder, a liquid nitrogen Dewar
vessel, pressure regulator, valves, a Data Acquisition (DAQ) system, pipes and brass fittings,
electric heater, flow meter and a test section consisting of a polyurethane-coated helical coil
with T-type thermocouples. Figure 1 shows a schematic of this. The test section was made
Figure 1. Schematic of experimental system setup.
480

of Oxygen-Free High thermal Conductivity (OFHC) copper tubing (UNS C10100) with an
internal diameter of 5/16 inch (0.0079 m). The test section had helical channels with major
diameter of 106 mm and was coated with polyurethane (PU) using a Chemical Vapor Depo-
sition (CVD) process. The thickness of the PU coating was 0.1 mm. The wall temperature
measurements were taken at five equidistant points along the curve length by using T-type
thermocouple wires. The leak of external heat into the test section was minimized by using a
thick layer of polyurethane foam (thermal conductivity of 0.02 W/m K, density of 11 kg/m3),
which is a type of expanded foam insulation. The test section was also covered by yarn and
then by nitrile rubber insulation. Liquid nitrogen was supplied to the test section through
½-inch 304-grade stainless-steel pipes and brass fittings from a 55 L Dewar vessel (TA-55)
made by IBP Co. Ltd. (Kolkata, India). The liquid nitrogen was pumped from the Dewar ves-
sel by external pressurization using a gaseous nitrogen cylinder having 47 L capacity, which
could supply a maximum pressure of 15 bar. The flow rate of liquid nitrogen was manually
regulated by a pressure regulator mounted on the external pressurization line. Initially, the
entire test section was purged with gaseous nitrogen. A bypass line was introduced to ensure
the entry of saturated liquid N2 into the test section. Temperature measurement of the test
section was done using T-type thermocouples connected to a Keysight 34972 A data acquisi-
tion/data logger switch unit with scanning frequency of 30 milliseconds, which was also con-
nected to a personal computer for data analysis. The average mass flux was measured using
a volume flow meter having an accuracy of 0.05 mm3/s at the exit line from the test section.
A mercury thermometer was mounted at the outlet to measure the average temperature of
the outgoing gas. A single-phase flow meter was provided to measure flow rate. To ensure
that a single-phase gas flow entered the flow meter, the outlet line was placed in a hot water
bath heated by electric heaters. A pressure gauge was provided to measure and regulate initial
pressure. Different flow control valves were provided to control the flow.
2.2 Procedure
Straight copper tube was first coated with polyurethane to a thickness of 0.1 mm. From
this, a helical test section with pitch angle of 8° and the above-mentioned dimensions was
prepared. The thermocouples were fixed on the surface of the test section circumferentially
(120° apart) at five equally spaced locations with three thermocouples in each section (in one
pitch length). The test section was covered first by yarn and then by nitrile rubber insulation.
Polyurethane foam was sprayed onto it for further insulation. The experiment was conducted
at two different inlet flow pressures of 6.89 kN/m2 and 8.61 kN/m2. The equivalent mass flow
rates were calculated to be 93 kg/m2 s and 116 kg/m2 s, respectively.
3.1 Temperature profile in chilldown process

The temperature profile determines how fast the cryogenic fluid can cool down the tube.
Comparison of temperature profiles is useful in identifying the chilldown effectiveness of
coated and uncoated surfaces. Three main boiling regimes, film boiling, transition boiling,
and nucleate boiling, are the focus while analyzing the chilldown phenomenon. The initial
phase is film boiling where because of a larger temperature difference between the cryogen
and the tube wall, the fluid undergoes rapid evaporation and the test section is filled entirely
with vapor. The point at which liquid comes into contact with the wall for the first time is
termed the Leidenfrost point and is the point of minimum heat flux. This represents the
initiation of the transition boiling regime. Because of the presence of liquid contact with
the wall, heat transfer greatly increases. This maximum heat transfer point is called the criti-
cal heat flux point, which signals the onset of the nucleate boiling regime. Because the heat
transfer coefficient is large, the nucleate boiling regime has a higher heat transfer rate and
results in faster cooling. On uncoated surfaces, the solid–fluid thermal resistance forms the
481

basis for the heat transfer at the interface. The vapor film developed immediately between
the wall and fluid during the initial flow period causes an increase in the time required for
chilldown because it takes time to break this boiling envelope. In the case of coated surfaces,
the low conductivity of the coating material results in a thermal gradient on either side caus-
ing a reduction in temperature between fluid–fluid interfaces, thus resulting in a reduction in
chilldown time.
3.2 Comparison of temperature profile between coated and uncoated surfaces

The experiments were conducted on two helical test sections with coated and uncoated
surfaces for two different mass flow rates. The average of all working thermocouples was
deemed as the chilldown time, and the results are shown in Figure 2. It can be seen that
the coated channel attains chilldown more quickly than the uncoated one; here, in the case
of coated channels, the time required to cover film boiling region is reduced to 28%. The
uncoated test section shows a gradual attainment of nucleate boiling while the coated test
section produces a sudden attainment of nucleate boiling, thereby resulting in a substantial
reduction in chilldown time. This significant chilldown time reduction is seen after 275 sec-
onds, that is, only after attaining the nucleate boiling regime. Figure 2 describes the vari-
ation of average surface temperature with time for the mass flow rate of 93 kg/m2 s. From
Figure 3, it is clear that, at the average chilldown temperature of 117 K, the uncoated tube
requires 352 seconds to reach this temperature, while the coated one requires just 298 sec-
onds, thereby resulting in a time saving of 54 seconds. Figure 4 illustrates the variation
of average surface temperature with time for the mass flow rate of 116 kg/m2 s. It can be
concluded that the significant chill down time reduction for this condition is only seen after
165 seconds, that is, only after the attainment of the nucleate boiling regime. However, for
the uncoated test section the transition occurs at around 72 seconds. The high specific heat
of the polyurethane coating could have led to the time delay in completing the transition
regime. From Figure 5, it can be seen that to reach an average chilldown temperature of
126 K, the uncoated tube requires 307 seconds, while the coated tube took 251 seconds,
thereby resulting in a time saving of 18%. The high specific heat of polyurethane keeps the
temperature difference higher for longer, which improves heat transfer and helps to reduce
the chilldown time compared to an uncoated coil. Figures 6 and 7 show the variation of
temperature along the tube from inlet to outlet at 200 and 300 seconds, respectively. At
200 seconds, the temperature varies between 118.9 K and 129.8 K for the coated tube and
between 123.7 K and 119.3 K for the uncoated one. Similarly, at 300 seconds, the tempera-
ture varies from 135.2 K to 135 K for the coated tube and 131.07 K to 141.60 K for the
uncoated tube. From Figure 7, it can be seen that the temperature towards the end of the
tube is increasing for both cases, which is due to reduced temperature difference between
Figure 3. Average surface temperature vs. time

Figure 2. Average surface temperature vs. time after 270 seconds for mass flow rate of 93 kg/m2 s.
for mass flow rate of 93 kg/m2 s.
482

Figure 5. Average surface temperature vs.
Figure 4. Average surface temperature vs. time time after 170 seconds for mass flow rate of
for mass flow rate of 116 kg/m2 s. 116 kg/m2 s.
Figure 6. Variation of temperature along tube Figure 7. Variation of temperature along tube
from inlet to outlet at 200 seconds. from inlet to outlet at 300 seconds.
the coil and the fluid. This in turn reduces heat transfer to the fluid, which causes a lower
reduction in temperature compared to the inlet temperature drop.
According to Cowley et al. (1962), chilldown time can be decreased by using a thin layer
of insulating material on the surface of the metal, and they also postulated that it can be
applied to the internal surface of cryopanels. The narrow region in which the maximum heat
transfer exists, after which the nucleate boiling begins, can be widened with the addition of a
low thermal conductivity coating material on the conducting surface. When enough fluid is
present, a thermal gradient is developed because of this layer that helps to attain the tempera-
ture in the critical maximum heat transfer region, resulting in the shortening of chill down
time. From our observations, with an increase in mass flow rate, the chilldown time decreases
as the quantity of fluid flowing through the tube increases. Because the film boiling regime
has a lower heat transfer rate than the nucleate boiling regime, we have focused our compari-
son on the nucleate boiling regimes of the coated and uncoated surfaces. On the basis of the
higher slope of the nucleate boiling regime of the coated tubes and the temperature profiles,
it can be inferred that heat transfer would be enhanced by coating the inner walls of the tube,
resulting in further reduction of chilldown time.
483

4 CONCLUSION AND SCOPE OF FUTURE WORK
In our experiment on the chilldown of polyurethane-coated and uncoated helical coils, it can be
concluded that polyurethane coating of the helical coil has increased the time required for tran-
sition from film boiling to nucleate boiling as a result of the high specific heat of polyurethane.
Because of this high specific heat, the temperature drop of the polyurethane-coated tube
takes longer than the uncoated one. This maintains the flow in a film boiling regime for
longer, but after transition the temperature drop was rapid and for a longer duration. Here,
the temperature drop of the polyurethane is considerably lower, which keeps the temperature
difference higher, causing higher heat transfer compared to the uncoated tube.
Because of the sudden increase in heat transfer after the transition, it was found that the
polyurethane-coated coil has a shorter chilldown time for different mass flow rates. These
results should prove useful in the design of transfer lines. Future work can employ different
combinations of materials for transfer lines and coating materials with lower conductivity.
The effectiveness of different geometries can also be investigated. An in-depth understanding
of this phenomenon can be obtained by considering the heat transfer coefficient variation in
the three flow regimes during the chilldown process.
ACKNOWLEDGMENTS
The authors would like to thank the Space Technology Laboratory of TKM College of
Engineering, and also the Technical Education Quality Improvement Programme Phase II
(TEQIP-II), promoted by the National Project Implementation Unit, Ministry of Human
Resource Development, Government of India, for their support.
REFERENCES
Allen, L. (1966). A method of increasing heat transfer to space chamber cryo panels. Advances in
Cryogenic Engineering, 11, 547–553.
Berger, S., Talbot, L. & Yao, L. (1983). Flow in curved pipes. Annual Review of Fluid Mechanics, 15,
461–512.
Chen, C.-N., Han, J.-T., Jen, T.-C. & Shao, L. (2010). Thermo-chemical characterstics of R134a flow
boiling in helically coiled tubes at low mass flux and low pressure. Thermochimica Acta, 512, 1–7.
Cowley, C., Timson, W. & Sawdye, J. (1962). A method for improving heat transfer to cryogenic fluid.
Advances in Cryogenic Engineering, 7, 385–390.
Dravid, A., Smith, K. & Merrill, E. (1971). Effect of secondary fluid motion on laminar flow heat trans-
fer in helically coiled tubes. AIChE Journal, 17, 1114–1122.
Fsadni, A.M. & Whitty, J.P. (2016). A review on the two-phase heat transfer characteristics in helically
coiled tube heat exchangers. International Journal of Heat and Mass Transfer, 95, 551–565.
Hardik, B. & Prabhu, S. (2017). Critical heat flux in helical coils at low pressure. Applied Thermal
Engineering, 112, 1223–1239.
Jensen, M.K. & Bergles, A.E. (1981). Critical heat flux in helically coiled tubes. Journal of Heat Transfer,
103, 660–666.
Johnson, J. & Shine, S. (2015). Transient cryogenic chill down process in horizontal and inclined pipes.
Cryogenics, 7, 7–17.
Leonard, K., Getty, R. & Franks, D. (1967). A comparison of cooldown time between internally coated
and uncoated propellant lines. Advances in Cryogenic Engineering, 12, 331–339.
Maddox, J. & Frederking, T. (1966). Cooldown of insulated metal tube to cryogenic temperature.
Advances in Cryogenic Engineering, 11, 536–546.
Prabhanjan, D.G., Raghavan, G.S.V. & Rennie, T.J. (2002). Comparison of heat transfer rates between
a straight tube heat exchanger and helically coiled heat exchangers. International Communications in
Heat and Mass Transfer, 29, 185–191.
Yuan, K., Chung, Y.J.N. & Shyy, W. (2008). Cryogenic boiling and two-phase flow during pipe chill-
down in earth and reduced gravity. Journal of Low Temperature Physics, 150, 101–122.
484

Effect of concentrated solar radiation on reduction of pressure

drop in oil pipeline
V.C. Midhun, K. Shaji & P.K. Jithesh

Department of Mechanical Engineering, Government Engineering College, Kozhikode, India
ABSTRACT: Oil transportation through pipeline is considered as an effective and eco-

nomical method. Heating is an effective method used for the reduction of crude oil viscosity
and pressure drop. In the present work, 3D steady state CFD analysis carried out in an oil
pipeline which is heated by applying concentrated solar radiation using a Parabolic Trough
Collector (PTC) where the pipeline acts as an absorber pipe. The analysis focuses on the effect
of concentrated solar radiation for reducing pressure drop in the oil pipeline. This reduction
in pressure drop helps to reduce the pumping power requirement. The influence of concen-
trated solar radiation and Reynolds number on pressure drop in heated oil pipeline is investi-
gated. Results from the CFD analysis indicates that considerable pressure drop reduction can
be achieved by heating the pipe with concentrated solar radiation using PTC.
1 INTRODUCTION
The solar energy is considered as the most essential, clean and inexhaustibly accessible renew-
able energy. Its two main applications can be classified into heating and generating electrical
energy. Solar energy utilisation has a great scope in the present situation. This work is related
to the utilization of solar energy to solve the transportation problem faces in oil pipelines.
It is observed that, heating is one of the usual methods used to reduce the crude oil viscos-
ity for reducing pressure drop when compared with other chemical treatment methods. In
(Midhun et al. 2015), authors mentioned a novel method of heating crude oil pipelines using
Parabolic Trough Collector (PTC) for reducing pumping power by applying concentrated
solar radiation on pipe surface. Here the authors tried to investigate the pressure drop in
heated oil pipeline and adiabatic pipe and its comparison is made to show the pressure drop
reduction. The hydrodynamic and thermal characteristics of the flow were also investigated
to explain the nature of flow and heat transfer inside the pipe. Also the relevance of a three
Figure 1. Oil pipeline at different conditions.
485

dimensional steady state numerical analysis of pipe line was discussed. But the effect of differ-
ent concentrated radiation and different flow condition were not investigated. In this paper, a
3D steady state CFD analysis on oil pipeline was carryout by applying different level of con-
centrated solar radiation its surface. Here a detailed analysis is carried out to find the effect of
different concentrated solar radiation (heat flux) and different Reynolds number flow in pres-
sure drop reduction in oil pipeline. In practical situation, there will be convective and radia-
tion losses occur from pipeline surface to surrounding. So a comparative analysis between the
heated oil pipeline with and without heat loss is also included. Through this the effect of heat
loss in pressure drop reduction is investigated in this work. Figure 1 shows the different condi-
tions that an oil pipeline confronts, which are analysed in this work.
2 METHOD
In this work, the main motive is to find the effect of Reynolds number of the flow and dif-
ferent level of concentrated solar radiation on pressure drop reduction in oil pipelines. Thus
it is necessary to conduct a single phase three dimensional analysis of the flow through an
oil pipeline to determine the pressure drop inside the pipeline along with hydrodynamic and
thermal characteristics of the flow. So this analysis mainly focuses on the impact of variation
of Reynolds number of the flow and heat flux on reduction of pressure drop in oil pipelines
heated by PTC. Numerical analysis is done by using CFD software tool ANSYS FLUENT
14.5.
The steady state equation for conservation of mass or can be written as follows:
∂

∂xi
( ρui ) = 0 (1)
Momentum Equation:
∂ ∂P ∂   ∂u ∂u j  2 ∂u 

∂xi
( )
ρui u j = − +
∂xi ∂xi
( µ + µt )  i +  − ( µ + µt ) l δ ij  + ρ gi
 ∂x j ∂xi  3 ∂xl 
(2)

Energy equation:
∂ ∂  µ µ  ∂T 

∂xi
( )
ρui u j =  + t  + SR
∂xi   Pr σ T  ∂xi 
(3)
The most frequently adopted turbulence model is the k – ε models. Transport equation for
Realizable k-ε Model:
∂ ∂  ∂k 

∂x j
(
ρ ku j =) ( µ + µt )
∂xi 
 + Gk + Gb − ρε −YM + Sk
∂x j 
(4)

∂ ∂  µt  ∂ε  ε2 ε

∂xi
( ρε ui ) =  µ + 
∂xi 
 + ρC1Sε − ρC2
σ ε  ∂xi 
+ C3 C3εGb + Sε (5)
k + ϑε k
where C1 = max  0.43, ηη+ 5  , η = S kε and S = 2Sij Sij

Gk and Gb corresponds the generation of turbulence kinetic energy due to the mean veloc-
ity gradient, and turbulence kinetic energy generation due to buoyancy, YM stands for the
contribution of the fluctuating dilatation in compressible turbulence to the overall dissipa-
tion rate. C1ε and C2 are the constants. σk and σε stands for turbulent Prandtl numbers for k
and ε. SR, Sk and Sε represents the user-defined source terms.
486

Single phase steady state fluid flow without gravitational effect is taken for the analysis. It
is assumed that heat flux around the pipe surface by direct and concentrated solar radiation
on the absorber tube is taken in approximated manner. Inner surface of the pipe is consid-
ered to be smooth.
2.1 Computational domain

Inner diameter and outer diameter of pipe are 0.02 m and 0.025 m. This small diameter
pipe is normally used in boilers for carrying furnace oil. The Figure 2 describes the practical
condition that an oil pipeline heated by concentrated solar radiation along with the heat loss
from pipe surface to surroundings. The computational domain is created by using ANSYS
Design Modeler. The fluid domain is for crude oil and solid domain for pipe material. The
pipe wall of 1 m from the inlet side is considered as adiabatic wall in all the three case. The
concentrated radiation is applied on the base of pipe surface by using a Parabolic Trough
Collector (PTC) of rim angle 160° where the pipeline acts as absorber pipe. In heated region
of the oil pipe, the concentrated solar radiation falls on the bottom surface of the pipe and
direct solar radiation falls on the top surface.
2.2 Mesh details

Meshing of the computational domain explained above was done by using ANSYS ICEM.
The total no of elements in the domain are 435774 with 454608 nodes. The fluid and solid
domain are created by using sweep method. Inflation is provided in the fluid domain closer
to the inside pipe wall so that fluid domain nearer to the pipe domain will be finely meshed
which is necessary to incorporate the effect of boundary layer. The surface of the pipe is
divided in to some parts for providing proper boundary conditions.
Average fluid outlet temperature is monitored with different levels of meshing for Reynolds
number 12000 and heat flux of 80000 W/m2. The variation of fluid temperature decreases
with increase in number of mesh elements and temperature approaches a value of 299.93 K.
So further increase in mesh elements have no considerable effect in temperature effect. The
temperature approaches a tolerance value of 0.001 K and the domain becomes grid inde-
pendent at 435774 mesh elements.

In this analysis fluid domain is Iranian Light Dead crude oil. Aluminium is taken as the pipe
material because of its reasonable mechanical and thermal properties. Properties of Alumin-
ium as ρ = 2719 kg/m3, cp = 871 J/kg K, k = 202.4 W/m K respectively. The heat capacity and
thermal conductivity of crude oil are 1887 J/kg K and 0.1483 W/m K. The density is taken
as a function of temperature:
Figure 2. Heating of oil pipeline by applying concentrated solar radiation on base surface (with heat loss).
487

ρ = −0.732T + 869.3 (6)
from (Tavakoli and Baktash 2012) where temperature T is in °C. The correlation used for
dynamic viscosity (µ) from (Sattarin et al, 2007) for light dead crude oil is,
b
e API
µ =a× (7)
API
where a = 0.00735T 2 − 4.3175T + 641.3572, b = −1.51T + 568.84 .

Here the dynamic viscosity µ is in centipoise (cP) and temperature T is in Kelvin (K).
2.4 Solution methodology

The governing equations were solved using the commercial software package of FLU-
ENT. The second order upwind scheme was employed for convective term discretization of
momentum, energy, turbulent kinetic energy and turbulent dissipation; whereas least square
cell based scheme for gradients and standard for pressure equation quantities at cell faces are
computed using a multidimensional linear reconstruction approach. SIMPLE algorithm was
chosen for pressure and velocity coupling. The solution was considered to be converged when
the difference was limited to the third decimal point for solution of velocity terms and the
sixth decimal point for energy and continuity solutions.
2.5 Boundary conditions

The pipe inlet is given with velocity inlet boundary condition with oil temperature of 298 K
and outlet with pressure outlet condition. The inlet velocity is taken on the basis of Reynolds
number at inlet with property values based on the temperature of the crude oil. The operating
pressure is taken as 101.3 kPa and a gauge pressure of 0 Pa is given as the outlet pressure. For
the analysis of insulated oil pipeline, the pipe wall is given with adiabatic wall condition whereas
in the case of heated pipeline, the pipe domain surface is split up in to two basic regions—
adiabatic region and heated region. 1 m length of pipe wall surface is given with adiabatic con-
dition. The remaining part of the pipe surface is split in to four regions (top and bottom surface
in each 1 m length). The top surface of 2 m length of pipe is provided with wall boundary
condition for applying the effect of direct solar radiation. The corresponding bottom portion
of the pipe is also provided with wall boundary condition to account the effect of concentrated
solar radiation. Heat generation (qg) equivalent to constant heat flux of 1000 W/m2 is given to
the top surface. In addition to that the effect of concentrated solar radiation is applied on the
pipe surface in the form of heat generation on pipe surface equivalent to the heat flux. This
equivalent value of heat generation is calculated by using the relation:
qg′′′ ×
π
4
( ( D + 10
0
−6
)
2
)
− D0 2 × L = q′′ × π D0 × L (8)
By using mixed type of wall boundary condition, the heating effect (by heat generation)
along with convection and radiation losses from the pipe surface can be considered. The
combined heat transfer coefficient for forced convection and natural convection loss from the
pipe to the surrounding can be calculated by a correlation against wind speed as from (Duffie
and Beckman 2013):
h0 = 5.7 + 3.8 uW (9)
where uw is the velocity of wind. In this work h0 is taken as approximately 10 W/m2.

The calculation for convection loss with atmospheric is based up on equation 9 assuming
the atmospheric temperature T∞ = 300 K, convection coefficient h0 = 10 W/m2 K. Radiation
488

losses from pipe surface to the sky are accounted by calculating the sky temperature by using
the relation given in (Sharma and Mullick 1991).
Tsky = 0.0552T∞1.5 (10)
from that Tsky = 286.82 K and the atmospheric temperature is taken as T∞ = 300 K. The emis-
sivity of sky εsky is calculated by using Trinity equation:
ε sky = 0.787 + 0.0028Tdp (11)
where εsky = 0.83 and 15.1°C as dew point temperature.
The main objective of the present work is to find the effect of Reynolds number of the flow
and different level of concentrated solar radiation on pressure drop reduction in oil pipelines.
The heat loss from the pipe to the surroundings is also considered in this analysis, so it can
be considered as almost a practical case. Due to this heat loss, naturally the effect of heating
in reducing pressure drop is less when compared to the ideal case (without heat loss) except
at higher level of heating. Here the hydrodynamic and thermal characteristics of the flow are
analysed and compared with ideal case in-order to find the effect of heat loss. The pipe line is
provided with non-uniform heating across the section, so average value of temperature, fric-
tion factor and Nusselt numbers at each section of the oil pipeline are considered here.
3.1 Effect of Reynolds number at constant heat flux (with heat loss)
Pressure distribution in the adiabatic pipe and heated pipe at different Reynolds number
can be compared from the Figure 3. The pressure drop in adiabatic pipe is higher than the
heated pipe for each Reynolds number. This pressure drop reduction in heated pipe is due to
decrease in viscosity.
Figure 3 shows the comparison of pressure variation with axial length at constant heat flux
and adiabatic condition under different Reynolds number. The effect of heat loss is not much
reflected in the reduction in pressure drop. The nature of pressure curve is similar to that
of the ideal case (without heat loss). The difference between the pressure drop in heated oil
pipe having thermal leakage and heated oil pipe at ideal case is very small. It is observed that
Figure 3. Pressure distribution along the heated Figure 4. Comparison of average friction fac-
pipe (at constant concentrated solar radiation of tor of the heated pipe along dimensionless lengt-
80000 W/m2) and adiabatic pipe under different hat concentrated solar radiation = 80000 W/m2
Reynolds number. under different Reynolds number.
489

pressure drop in ideal case is higher when compared with pressure drop in heated pipe with
heat loss. Because, here the increase of friction factor due to increase in velocity gradient at
heated pipe wall near to the outlet is lower than the ideal case.
The variation of friction factor (f) with dimensionless length (X/D) in a heated pipe at dif-
ferent Re is shown in the Figure 4. Friction factor will be higher at the entrance region due
to higher velocity gradient and attains a constant value at fully developed region but here the
pipe is heated by direct and concentrated solar radiation, so the viscosity and density of fluid
are reduced. After the sudden decrease in friction factor there is a gradual increase in fric-
tion factor along the axial length in the heated region. This rise in friction factor is due to the
increase in velocity gradient at the pipe wall. Decrease in fluid density near to the heated pipe
wall leads to increase in fluid velocity near the pipe wall. Thus the velocity gradient increases
and overcomes the effect of decrease in viscosity.
Figure 5 shows the pressure drop variation with Reynolds number in heated pipe and adi-
abatic pipe where the concentrated solar radiation (heat flux) is maintained as 80000 W/m2.
From the plot it is observed that the difference between the pressure drop in adiabatic pipe
and heated pipe is increasing with increase in Reynolds number. While increasing Reynolds
number there will be a decrease in friction factor. Pressure drop difference between adiabatic
pipe and heated pipe is increasing with increase in Reynolds number. Thus at higher Rey-
nolds number this heating method is more effective in reducing the pressure drop for this
corresponding heat flux.
3.2 Effect of different heat flux (concentrated solar radiation) at constant Reynolds number
In this analysis the Reynolds number at the inlet is maintained at 3000 while the concentrated
radiation (heat flux) is varied around the bottom of the oil pipe surface. Here also the heat
loss from the pipe is considered.
Sudden decreases in friction factor due to decrease in viscosity by heating can be observed
from Figure 6, but after that f attains a very small rise along the axial length. This is because
the density of fluid near to the wall decreases further while the viscosity variation is very
small. The velocity of fluid near to the wall increases because of decrease in density. This
increase in fluid velocity leads to the increase in wall shear stress because of high velocity
gradient at the wall surface. Thus the friction factor began to increase gradually up to outlet
of the pipe. It leads to an adverse effect on pressure drop reduction with increase in heat flux
which is analysed further in detail by obtaining plot for pressure drop versus heat flux.
Figure 7 shows the comparison of pressure drop between oil pipes at ideal case and with
thermal leakage at constant Re under different heat flux. Pressure drop in heated pipe with-
out heat loss and with heat loss are plotted against heat flux. The pressure drop increase is
Figure 5. Comparison of pressure drop variation with

Reynolds number in adiabatic pipe and heated pipe. Figure 6. Comparison of average friction
factor along the pipe at Re = 3000 and differ-
ent concentrated solar radiation (heat flux).
490

Figure 7. Comparison of pressure drop Figure 8. Comparison of Nussult number from
variation with heat flux (Re = 3000). CFD analysis with Gnelinskie’s correlation.
due to the increase in friction factor due to increase in velocity gradient. This friction factor
increases with heat flux after a particular value of heat flux. This may depend on the length
of heating, heat loss and fluid properties. It is clear that the heat loss has some effect in the
pressure drop reduction. The pressure drop curve is shifted towards right side. Pressure drop
in ideal case (without heat loss) is lesser than the pressure drop in pipe with heat loss up to
certain limit of heat flux after that the pressure drop in pipeline with heat loss is found to be
higher than ideal case.
3.3 Validation
Average Nusselt number obtained from the CFD analysis are used for validation purpose.
Nusselt number is compared with the value obtained from Gnelinskie’s correlation. Here
the Re = 3000 at inlet and the concentrated solar radiation as 40000 W/m2 with heat loss
condition.
Fluid properties were calculated based on this bulk mean temperature. While the average
Nusselt number for oil pipeline with heat loss obtained from analysis agrees with Gnelinski’s
correlation (Eq. 11) with an error of 6.48%. This may be due to the effect of severe property
variation and large difference between wall temperature and fluid temperature.
Nu =
( f
8 ) ( Re 1000) Pr
− ×
, (11)
1 + 12.7 ( f 8 ) (Pr − 1)
0.5
2
3
4 CONCLUSIONS
The utilisation of PTC for pressure drop reduction in oil pipelines by applying concentrated
solar radiation on the pipe surface, in order to reduce pumping power was analysed. Pressure
variation in heated oil pipeline was compared with adiabatic oil pipeline for corresponding
Reynolds number of flow to determine the effect of heating in pressure drop reduction. The
analysis shows that with increase in Reynolds number, the effect of heating in pressure drop
reduction is getting significantly increased. Heating effect is found to be more effective in higher
Reynolds number. The pressure drop versus different concentrated solar radiation was plotted.
Also the effect of pressure drop reduction varies with different concentrated solar radiation in a
peculiar manner. At first the pressure drop gets reduced with increase in heat flux but after that
pressure drop curve showed a reverse trend due the increase in friction factor.
491

NOMENCLATURE
D Diameter (m)
f Friction Factor
h Convective heat transfer coefficient (W/m2 K)
k Thermal conductivity of fluid (W/mK)
Nu Nusselt number
P Pressure (Pa)
Pr Prandtl number
qg Volumetric heat generation rate (W/m3)
q” Rate of heat transfer per unit area (W/m2)
Re Reynolds number
T Temperature (K)
u Velocity (m/s)
X Axial length (m)
Greek symbols
ε Emissivity
µ Dynamic viscosity (Ns/m2)
ν Kinematic viscosity (m2/s)
ρ density (kg/m3)
REFERENCES
Cengel Y.A. 2013. Heat Transfer A Practical Approach. Cambridge: Cambridge University Press.
Duffie, J.A., & Beckman, W.A. 2013. Solar Engineering of Thermal Processes. Hoboken, NJ, USA: John
Wiley & Sons, Inc.
Forristall, R. 2003. Heat Transfer Analysis and Modeling of a Parabolic Trough Solar Receiver Imple-
mented in Engineering Equation Solver. Golden, Colo.: National Renewable Energy Laboratory,
NREL/TP; 550-34169.
Hart, A. 2014. A Review of Technologies for Transporting Heavy Crude Oil and Bitumen via Pipelines.
Journal of Petroleum Exploration and Production Technology 4(3): 327–36.
Mammadov, F.F. 2006. Application of Solar Energy in the Initial Crude Oil Treatment Process in Oil
Fields. Journal of Energy in Southern Africa 17(2): 27–30.
Martínez-Palou, R., Mosqueira, M. de L., Zapata-Rendón, B., Mar-Juárez, E., Bernal-Huicochea, C.,
de la Cruz Clavel-López, J., & Aburto, J. 2011. Transportation of Heavy and Extra-Heavy Crude Oil
by Pipeline: A Review. Journal of Petroleum Science and Engineering 75(3–4): 274–82.
Midhun V.C., Shaji, K. & Jithesh, P.K. 2015. Application of Parabolic Trough Collector for Reduction
of Pressure Drop in Oil Pipelines. International Journal of Modern Engineering Research 5(3): 40–48.
Price, H., Lüpfert, E., Kearney, D., Zarza, E., Cohen, G., Gee, R., & Mahoney, R. 2002. Advances in
Parabolic Trough Solar Power Technology. Journal of Solar Energy Engineering, 124(2), 109.
Matthew Roesle, Volkan Coskun, & Aldo Steinfeld. 2011. Numerical Analysis of Heat Loss From a
Parabolic Trough Absorber Tube With Active Vacuum System. Journal of Solar Energy Engineering
133(3): 31015.
Saniere, A., Hénaut, I. & Argillier, J.F. 2004. Pipeline Transportation of Heavy Oils, A Strategy, Eco-
nomic and Technological Challenger: Oil and Gas. Science and Technology-Rev. IFP 59(5): 455–466.
Sattarin, M., Modarresi, H. & Teymori, M. 2007. New Viscosity Correlations for Dead Crude Oils.
Petroleum & Coal 49(2): 33–39.
Sharma, V. B & Mullick, S.C. 1991. Estimation of Heat-Transfer Coefficients, the Upward Heat Flow,
and Evaporation in a Solar Still. Journal of solar energy engineering 113(1): 36–41.
Tavakoli, A., & Baktash, M. 2012. Numerical Approach for Temperature Development of Horizontal
Pipe Flow with Thermal Leakage to Ambient. International Journal of Modern Enginering Research
2(5): 3784–94.
492

Effects of air jet on bluff body stabilized flame: Validation by

simulation
S. Parvathi, V.P. Nithin, S. Nithin, N. Nived, P.A. Abdul Samad & C.P. Sunil Kumar
ABSTRACT: A burner having a conical bluff body with a central air injector is consid-
ered. In this paper, the effects of the central air jet on the heat load of the bluff body are
investigated. The flame structures and the flame blowoff temperatures were compared with
corresponding simulated outcomes. Simulation results show that the considerable reduction
in the heat load to the bluff body by the central air jet determined experimentally is quite
valid. Thus the problem caused by the high heat load in practical applications has a solu-
tion. The addition of central air jet alters the flame structures and blowout temperatures, as
shown in simulation as well as experiment. Various blowout behaviors caused by the air jet
observed experimentally also match those that were simulated. It is evident from simulation
and experimentation that the center air injection could cool down the bluff-body. However,
the flame stability could not be accomplished.
1 INTRODUCTION
In a multitude of applications, such as afterburners, gas turbine combustors, industrial fur-

naces and heat recovery steam generators, flame stabilization in premixed fuel–air streams is
of technological interest (Chaparro & Cetegen, 2006).
For stabilising a diffusion flame, the bluff-body with a central jet is used. This is because
the use of a bluff body flame holder improves the mixing properties and also increases
the control over combustion. The recirculation of hot gas behind a bluff-body enhances
mixing and reignites gas mixtures (Guo et al. 2010). But the studies on the flame hold-
er’s temperature distribution were minimal. Bluff-body stabilized premixed flame has been
studied before in a few works by Zukoski & Marble (1955a), Zukoski & Marble (1955b),
Longwell (1953), Longwell et al. (1953), Wright (1959) and Pan et al. (1991). The effects
of some properties and geometry of the bluff body on the blowoff performance of flame
holders supplied with mixtures of gaseous propane and air were summarized in Lefebvre
et al. (2010). Shanbhogue et al. (2009) and Chaudhuri et al. (2009) investigated the dynamics
of two-dimensional bluff-body stabilized flames. Roquemore et al. (1986) tested the reac-
tion behavior of flows in an axisymmetric bluff-body burner. Esquiva-Dano et al. (2001)
concluded that six regimes of non-premixed bluff-body stabilized flames existed, based on
his experiment. Tang et al. (2013) studied the effects of the Reynolds number of central
fuel and annular air jet on the flame extinction and its structure. Clearly none of the exist-
ing works the effects of central air jet on bluff body stabilized flames by simulation. The
high heat load and temperature of the bluff-body surface were the challenges of the bluff
body in many practical applications. To understand the influence of the central jet on the
bluff-body surface temperature and the premixed flame stabilization, few efforts have been
identified. In this work, the investigated effects of central air jet on the bluff-body stabilized
premixed methane-air flame is compared with the results arrived at through simulation and
thus validated. The flame blowoff temperatures, bluff-body surface temperature and flame
structures were mainly considered.
493

Figure 1. Schematic of the nozzle (all dimensions are in mm).
2 METHODOLOGY
The schematic of the conical bluff-body burner is shown in Figure 1. After the literature sur-
vey, the simulation model was decided. Since the bluff body burner was symmetrical, a 2-D
simulation was chosen. The model for the same was created in Ansys Fluent software, which
is based on the finite volume method. The general transport equations for mass, momentum,
energy etc. are applied to each cell and discretized. All equations are then solved to render
the flow field. As in the experiment carried out by Tong et al. (2017), a 45° conical bluff-
body was placed in the center of the burner. The inner diameter of the circular pipe for the
methane-air flow is 30 mm. The bluff-body has top and inner diameters of 14 mm and 4 mm
respectively. The thickness of the pipe wall is 2 mm. Premixed methane-air is fed through the
annular channel. Air is injected through the central pipe. The mass flow rate of the methane-
air mixture is carried by varying the equivalence ratio of the same. The equivalence ratio is
set between Φannular = 0.64 to blowoff limit.
The boundary conditions are applied as per the experiment. The velocities are varied
accordingly. The combustion equation of methane is chosen as given in the Fluent software.
3 APEX TEMPERATURE VS. TIME
As cited in the literature (Euler et al. 2014), the temperature distribution is highest at the
center of the bluff-body. Tapex is taken as the temperature at the apex of the bluff-body sur-
face. For the boundary condition, the emissivity (ε) of the bluff-body is set as 0.58 (ε of
stainless steel varies 0.54 to 0.63). Taking central air jet velocity U-jet = 0, annular velocity
U-annular = 2.77 m/s and the annular flow equivalence ratio Φannular = 0.64, the temperature at
the apex of the injection hole is approximately 480 K, which is referred as T0 as the reference.
Fluent is used for simulation throughout the study. Figure 2 shows temperature changes of
Tapex when shutting down the fuel supply in simulation as well as in the experiment.
The fuel supply is shut down by reducing the mass flow rate of the methane-air flame
gradually to zero. Because of the weak flame attached to the bluff-body, the temperature
decreases with a sharp slope at first. Thereafter, due to the heat convection to the environ-
ment, when the effect of the flame completely disappears, the surface temperature changes
slowly over time. After the blowoff, the rate of decrease of temperature is less than 3 K/s in
both cases. This temperature is taken as Tapex.
In the simulation as well as the experimental result, the temperature before shutting down
the fuel supply is 513 K at 1 sec. After shutting down fuel supply, at 0 sec, the temperature
drops down to 481 K, in both cases.
In both cases, we have selected the temperature of the bluff body surface at times within
1 second after the flame is totally blown off. Thus the simulated pattern matches the experi-
mental results.
494

Figure 2. Changes of apex temperature when shutting down the fuel supply in experiment as well as
simulation.
4 GOVERNING EQUATIONS USED IN SIMULATION
∂ ∂ ∂
Continuity: ( ρu ) + ( ρv ) + ( ρw ) = 0
∂x ∂y ∂z
∂ ( ρuu ) ∂ ( ρuv ) ∂ ( ρuw ) ∂p ∂τ xx ∂τ xy ∂τ xz
Momentum: + + =− + + + + ρ fx
∂x ∂y ∂z ∂x ∂x ∂y ∂z
∂ ( ρvu ) ∂ ( ρvv ) ∂ ( ρvw ) ∂p ∂τ ∂τ ∂τ
+ + = − + yx + yy + yz + ρ f y
∂x ∂y ∂z ∂y ∂x ∂y ∂z

∂ ( ρwu ) ∂ ( ρwv ) ∂ ( ρww ) ∂p ∂ τ ∂τ ∂τ
+ + = − + zx + zy + zz + ρ fz
∂x ∂y ∂z ∂z ∂x ∂y ∂z

ρ – density, P – pressure, ζ – shear stress, f – body force, U,v,w – velocity components in
x,y,z directions.
∂ ( ρuh ) ∂ ( ρvh ) ∂ ( ρwh ) ∂ ( k∂T ) ∂ ( k∂T ) ∂ ( k∂T )

Energy: + + = + + +
∂x ∂y ∂z ∂x 2 ∂y 2 ∂z 2
 ∂  ∂Y  ∂  ∂Y  ∂  ∂Y  
∑i  ∂x  hi ρDi ,m ∂xi  + ∂y  hi ρDi ,m ∂yi  + ∂z  hi ρDi ,m ∂zi   − ∑ i hi Ri
 
Species: ∂ ( ρYi u ) + ∂ ( ρYiv ) + ∂ ( ρYi w ) = −  ∂  ρ D ∂Yi  + ∂  ρ D ∂Yi 

∂x ∂y ∂z  ∂x  i,m ∂x  ∂y  m ∂y 
i,m

∂  ∂Yi  
+ ∂z  ρ Di,m ∂z   + Ri
h – Enthalpy of species, k – Thermal conductivity, T – Temperature, Y – Mass fraction of
species, R – Production or consumption of species, D – Mass diffusivity.
5.1 Bluff-body surface temperature

Taking the premixed annular flow velocity Uannular = 2.77 m/s and equivalence ratio
Φannular = 0.64, while changing Ujet/Uannular = 0∼8.8, Tapex/T0 variation in simulation, the results
495

compared with the experimental results. As shown in Figure 3, the temperature of the inner
apex of the bluff-body surface varies with the injection of the central air jet.
Figure 3 depicts that in all three cases, namely simulation, experimentation and steady
condition, the temperature of the bluff body surface drops to less than 81% of T0, even with
a small amount injection of the central air jet. In the experiment, the temperature ratio comes
back to approximately 85% with an increase of the central air flow, which clearly matches
with the 86% achieved in simulation. Deviation to 95% in steady conditions is noteworthy.
The annular flow near the bluff-body dominates the local flow structures. That is why the
central air jet with Ujet/Uannular ≈ 1 causes the peak cooling effect.
A layer of air is formed above the bluff-body surface when central air jet is introduce. Thus the
heat convection from the flame to the bluff-body is prevented. When the velocity ratio is increased
to 1, neither flow dominates, making the effect of cooling weaker. Thus, to reduce the heat load to
the bluff-body, the central injection of air can be practically employed, taking into account Tapex.
Obviously with regard to bluff body temperature, simulation results validate the achieved
experimental outcomes.
5.2 Flame blowoff temperature

To evaluate the performance of the bluff-body stabilized flame, bluff body face temperature vari-
ation with respect to face radial distance is considered (Figure 4). It is seen that variation of blow-
Figure 3. Comparison of effects of central air jet on the temperature of bluff-body surface in experi-
ment. simulation and steady state.
Figure 4. Blowoff temperature distribution with respect to bluff body face radial distance.
496

off temperature occurs between 2.5 mm and 3.25 mm face radial distance in all velocity ratios
Ujet/Uannular. The highest blowoff temperature occurs without the central air jet. As the velocity
ratio Ujet/Uannular increases, the blowoff face temperature also increases. For all velocity ratios, the
maximum face temperature occurs at a face radial distance of 2.9 mm. When there is no central
air jet, the maximum blowoff face temperature reaches 495 K. Furthermore it becomes 393 K,
416 K, 438 K and 440 K as the velocity ratios change to 1, 2.46, 4.87 and 8.8 respectively.
Taking Uannular ≈ 1.85 m/s and Φannular = 0.64, steady state face temperature distribution with
respect to face radial distance are also found (Figure 5). Obviously for both cases, the overall
flame blowoff temperature increases with the introduction of the central air jet, which is one
of the main drawbacks.
At the beginning, the increase in flame blowoff temperature is due to the annular domi-
nated flow structure. A fresh air layer is formed by the central air jet upon the bluff body
surface. When the annular flow velocity is higher, the recirculation zone becomes small. This
results in detached flame from the bluff-body, which makes it easy to blowoff.
When the velocity ratio is equal to 1, the flame may get reattached to the bluff-body’s sur-
face. Also, when central air jet dominates the flow field, it results in a linear increase in the
flame blowoff temperature.
5.3 Flame structures

Taking Uannular = 1.85 m/s and Φannular = 0.64, Figure 6 shows the contour of temperature along the
flame structures. Evidently the flame structures change with the injection of the central air jet.
Figure 5. Steady state temperature distribution with respect to bluff body face radial distance.
Figure 6. Contour of temperature (K).
497

Figure 7. Temperature vs. axial distance.
With conditions of Ujet/Uannular = 0, 1, 2.46, 4.87 and 8.8 (from left to right), variations of
temperature in K with respect to axial distance are shown in Figure 7.
From Figure 6, it is obvious that the temperature downstream of the bluff-body is the high-
est without the introduction of the central air jet. When the central air jet is injected, it creates
a layer separating the main heat release zone and the flame, thereby appearing to get attached
to the surface of the bluff-body. The fresh cold air from the central jet fills the non-luminous
recirculation zone. With the addition of central air jet, the fuel air ratio decreases creating
leaner. The flame becomes weaker due to the reduction in size of the heat release zone. That
is why the bluff-body temperature with central air jet is smaller than T0. With small velocity
ratios of Ujet/Uannular, the annular flow dominates the flow field making the temperature with
Ujet/Uannular ∼ 1 the lowest, as is evident in Figure 7. When the recirculation zones downstream
of the bluff-body are dominated by the annular flow, to avoid the flame getting attached to
the bluff-body, the small amount of central air jet may form a fresh air layer. Also the heat
convection of burnt products to the bluff-body is reduced by the cold central air jet layer.
As is evident from Figure 7, the peak temperatures occur at axial distances of 150 mm,
200 mm, 260 mm and 300 mm as velocity ratios change from 1, 2.46, 4.87 and 8.8 respec-
tively. This means that the heat releasing zone becomes larger and travels to farther distances
with increases in central air jet velocity. Obviously in the absence of central air jet, the peak
temperature of 1,550 K occurs at 50 mm distance and the heat releasing zones cling together
in small region.
6 CONCLUSION
In this paper, the effects of central air jet on the bluff body stabilized premixed methane-air
flame, namely bluff-body surface temperature, flame blowoff temperature and flame struc-
tures are studied by simulation which, in turn, resulted in the validation the corresponding
experimental outcomes given by Tong et al. In both cases, it can be seen that the central air
jet reduces the heat load on the bluff-body surface. But on further addition of the air jet, the
flame becomes unstable. The flame blows off easily with the central air jet. Variation of apex
temperatures with different velocity ratios, variation of flame blowoff temperature with face
radial distance and variation contour of temperature in flame structure with different veloc-
ity ratio are thus validated by simulation study.
498

REFERENCES
Chaparro, A.A. & Cetegen, B.M. (2006). Blowoff characteristics of bluff-body stabilized conical
premixed flames under upstream velocity modulation. Combustion and Flame, 144(1), 318–335.
Chaudhuri, S. & Cetegen, B.M. (2009). Blowoff characteristics of bluff-body stabilized conical premixed
flames in a duct with upstream spatial mixture gradients and velocity oscillations. Combustion Sci-
ence and Technology, 181(4), 555–569.
Esquiva-Dano, I., Nguyen, H.T. & Escudie, D. (2001). Influence of a bluff-body’s shape on the stabiliza-
tion regime of non-premixed flames. Combustion and Flame, 127(4), 2167–2180.
Euler, M., Zhou, R., Hochgreb, S. & Dreizler, A. (2014). Temperature measurements of the bluff
body surface of a Swirl Burner using phosphor thermometry. Combustion and Flame, 161(11),
2842–2848.
Guo, P., Zang, S. & Ge, B. (2010). Technical brief: predictions of flow field for circular-disk bluff-body
stabilized flame investigated by large eddy simulation and experiments. Journal of Engineering for
Gas Turbines and Power, 132(5), 054503.
Lefebvre, A.H. & Ballal, D.R. (2010). Gas turbine combustion. CRC Press.
Longwell, J.P., Frost. E.E. & Weiss, M.A. (1953). Flame stability in bluff body recirculation zones.
Industrial & Engineering Chemistry, 45(8), 1629–1633.
Longwell, J.P. (1953). Flame stabilization by bluff bodies and turbulent flames in ducts. Symposium
(International) on Combustion. Elsevier, 4(1).
Pan, J.C., Vangsness, M.D. & Ballal, D.R. (1991). Aerodynamics of bluff body stabilized confined tur-
bulent premixed flames. ASME 1991 International Gas Turbine and Aeroengine Congress and Exposi-
tion. American Society of Mechanical Engineers.
Roquemore, W.M., Tankin, R.S., Chiu, H.H. & Lottes, S.A. (1986). A study of a bluff-body combustor
using laser sheet lighting. Experiments in Fluids, 4(4), 205–213.
Shanbhogue, S.J., Husain, S. & Lieuwen, T. (2009). Lean blowoff of bluff body stabilized flames: Scal-
ing and dynamics. Progress in Energy and Combustion Science, 35(1), 98–120.
Tang, H., Yang, D., Zhang, T. & Zhu, M. (2013). Characteristics of flame modes for a conical bluff body
burner with a central fuel jet. Journal of Engineering for Gas Turbines and Power, 135(9), 091507.
Wright, F.H. (1959). Bluff-body flame stabilization: blockage effects. Combustion and Flame, 3,
319–337.
Tong, Y., Li. M., Thern, M., Klingmann, J., Weng, W., Chen, S. & Li, Z. (2017). Experimental investi-
gation on effects of central air jet on the bluff body stabilized premixed methane-air flame, Energy
Procedia, 107, 23–32.
Zukoski, E.E., Marble, F.E. (1955a). The role of wake transition in the process of flame stabilization on
bluff bodies. AGARD Combustion Researches and Reviews, 167–180.
Zukoski E.E., Marble F.E. (1955b). Gas dynamic symposium on aerothermochemistry. Northwestern
University, Evanston, IL.
499

Identification of energy-intensified equipment

for reliability analysis
C. Sanjeevy & Jacob Elias

Department of Mechanical Engineering, School of Engineering, CUSAT, Cochin, India
ABSTRACT: A methodology is developed for the identification of Energy-Intensified

Equipment (EIE) for reliability analysis in the chemical processing industry. There are several
methods based on classification that can be used for identifying such equipment, such as
Always Better Control/ABC, Vital/Essential/Desirable (VED), Scarce/Difficult/Easily avail-
able (SDE), High/Medium/Low (HML), and Fast/Slow/Non-moving (FSN), but these selec-
tive inventory control methods do not indicate the criticality of an item. The damage grounds
from failure and the failure modes to plan an optimum maintenance program. The method
is applicable in understanding the equipment in the operational phase where there is only
limited data available. When available data is scarce or generic, critical data is retrieved from
some related selective inventory data banks. In this method, based on physical factors, the
situation under which the equipment is working, such as external/internal load/pressure, is
used in modeling the equipment. Pareto’s 80/20 principle is employed to identify its criticality
and calculate its risk factor. The current methodology applies criticality importance analysis
and criticality allocation to optimize the maintainability correlated with Reliability-Centered
Maintenance (RCM) models. Evaluating the reliability of life-threatening equipment in
reverse engineering of the (competitive) operational phase is one of the applications of this
method. As a case study, EIE is used for assessment of the proposed method and the results
identify the equipment and sub-systems that are critical elements from a reliability and main-
tenance perspective. A benchmark of the results indicates the effectiveness and quality of the
method in identification of energy-intensified equipment for reliability analysis.
Keywords: Energy-intensified equipment, reliability analysis, reliability-centered

maintenance
1 INTRODUCTION
For any item of equipment that has a vital role in a production sequence, a hazard in such
critical equipment will affect the entire production and create greater damage, so requires sig-
nificant care. The identification of Energy-Intensified Equipment (EIE) is one of the crucial
phases of Reliability-Centered Maintenance (RCM), through combination of quantitative
analysis with qualitative analysis (Barabady & Kumar, 2008). This research has been carried
out in a chemical company called Travancore Cochin Chemicals (TCC) Limited, situated in
Ernakulam district, Kerala, India. The chemical processing industry plays an essential role in
the manufacture of many chemicals, such as caustic soda, sodium chloride, chlorine, sulphu-
ric acid, hydrochloric acid and bleaching powder. There are more than 600 types of equip-
ment involved in their production. The industry provides a tremendous variety of materials
to other manufacturers, such as textiles, rayons, plastics, aluminum, detergents, drugs, ferti-
lizers, food preservatives, and paper-producing industries. It also produces chemical prod-
ucts that benefit people directly. Several changes in equipment have been taking place in the
processing of chemicals, and chemical processing industries in India are facing certain chal-
lenges that need to be addressed for their survival in the era of globalization. This analysis
501

focuses on the implementation of a rational methodology to detect reliability-related issues in
the chemical processing industry. An intense literature survey has been conducted in relation
to the equipment-related problems of chemical processing industries. The survey makes clear
that several aged chemical processing industries face severe problems in equipment identifi-
cation. The most significant problem is non-adoption of innovative technologies in produc-
tion. Because the technology in use therefore remains ancient, criticality-related difficulties
become common. The investigation of criticality and its combination with the reliability of
equipment are used to identify the related problems of analytical research.
2 LITERATURE REVIEW
This literature review is a brief review of research conducted in the areas of reliability and
maintenance programs of equipment, in order to identify the equipment and tools to be
adopted for the present study. Initially, the literature reveals several methods that can be used
to identify equipment, based on classifications such as ABC/Always Better Control, Vital/
Essential/Desirable (VED), Scarce/Difficult/Easily available (SDE), High/Medium/Low
(HML), and Fast/Slow/Non-moving (FSN). methods have not found the criticality, mutila-
tion causing from failure and the failure modes as well as to plan an optimum maintenance
program (Sanjeevy & Thomas, 2014).
Today, the methods most commonly applied in this field are Failure Mode Effect and
Criticality Analysis (FMECA) and Reliability-Centered Maintenance (RCM). Reliability
can be expressed as the possibility of process or equipment which perform its function or
task under stated environment for a definite surveillance period. Reliability analysis meth-
ods have been increasingly accepted as typical tools for the development and management
of regular and intricate processing methods since the mid-1980s. The occurrence of failure
cannot be prevented completely, but it is important to reduce both its chance of happen-
ing and the impact of failures when they occur (Barabady & Kumar, 2008). To sustain
the intended reliability, availability, and maintainability features and to attain expected
performance, a valid maintenance plan is essential. Both corrective and preventive mainte-
nance have direct consequences on the reliability of equipment and, consequently, its per-
formance. Hence the identification of energy-intensified equipment is critical for reliability
evaluation procedures. To overcome the limitations of the traditional classifications such
as the selective inventory controls of ABC, VED, SDE, HML, and FSN, and evaluation
using FMECA methodology, the present methodology is adopted. Ben-Daya and Raouf
(1996) noted that the economic model proposed by Gilchrist (1993) addresses a problem
that differs from the problem FMECA is intended to address. They combined the expected
cost model proposed by Gilchrist with their improved Risk Priority Number (RPN) model
in order to provide a quality improvement technique at the production stage. They also
confirmed that if the assessment of the factor scores on a 1 to 9 scale is not appropriate
then the treatment of identical significance is not practical. According to their model, the
probability of an event should be more significant and their model suggests the probability
of an event (with scale 1–9) is increased to the power of 2 (Tang et al., 2017; Puthillath &
Sasikumar, 2012).
There are two kinds of criticality analysis: quantitative and qualitative. To use the quan-
titative criticality analysis method, the investigation group has to identify the dependabil-
ity/unpredictability for every element, in a specified working period, to recognize the part
of the element’s unpredictability that can be attributed to each probable failure mode, and
rate the possibility of loss (or severity) that will result from each failure mode that can
occur (Sachdeva et al., 2009). Several authors make use of fuzzy set theory to tackle uncer-
tainties in maintenance decision-making, Chang et al. (1999) argued for the use of gray
theory to obtain critical valuations. The use of fuzzy logic theory for maintenance-critical
inquiry is also suggested in the literature (Eti et al., 2006; Teng & Ho, 1996; Jayakumar &
Asgarpoor, 2004).
502

3 RESEARCH METHODOLOGY
In reality, all items of equipment cannot be controlled with equal attention. An effective criti-
cal equipment identification calls for an understanding of the nature of care. Some equip-
ment may be very important while some is too small or too unimportant to call for a rigorous
and intensive mechanism. Criticality analysis means variances in the method of control from
equipment to equipment, founded on the basis of physical factors. The criterion used for
this purpose may be criticality, risk, maintenance difficulties, or something else. Controlling
the area of operation for good performance involves the time, money and effort required to
conduct operations that take less time and avoid sudden damage. Therefore, to achieve this
objective it need not be necessary to control the entire area of operations but only that area
of operation that is not controlled and is likely to cause damage. Thus, criticality analysis
means selecting the areas of mechanism so that the required objective is achieved as early as
possible without the loss of time that would be involved in taking care of the full area.
3.1 Critical analysis

When monitoring equipment, a company may be using many types in various numbers and
configurations. Equipment monitoring needs rational and clear classification. Thus, equip-
ment identification is needed to ensure objectivity in monitoring that equipment for mainte-
nance processes. The critical ranking number of a system or piece of equipment is a measure
of the system’s or equipment’s influence on the production when the system or equipment
fails to function, no matter how frequently the malfunction occurs. Clear critical ranking
rules are needed to help in allocating critical rankings to systems or equipment through the
analysis. The procedure is established by considering the combined scores for all evaluation
criteria. The critical ranking numbers, number range, and the model for assigning weighted
scores to the systems or equipment under assessment are defined before conducting the anal-
ysis. Critical ranking numbers are allocated to systems or equipment based on the procedures
developed. This is accomplished by comparing the equipment’s criteria with the important
factors in the critical ranking number rules. If the equipment matches the rules, the equip-
ment is assigned that critical ranking number. The equipment is always allocated the highest
critical ranking number that it scores. The 24 most appropriate critical factors selected for
equipment grouping are as listed below:
1. Percentage of utilization of machine
2. External/internal load/pressure acting on the equipment
3. Availability of machine
4. Weight of machine
5. Availability of substitute equipment
6. Impact on other equipment of breakdown
7. Cost of machine
8. Power consumption of machine
9. Age of equipment
10. Speed of repair
11. Equipment material
12. Availability of spares
13. Manufactured or purchased externally
14. Quality of work performed on machine
15. Ease of purchase
16. Effect of corrosion
17. Machine handling
18. Repaired or disposed of
19. Re-usability
20. Reliability
21. Equipment life cycle
503

22. Recyclability
23. Environmentally friendly
24. Maintenance history.
From these 24 basic factors, the nine most common factors suitable for the critical analysis
are selected and a questionnaire is prepared, based on a five-point “Likert” scale. This is con-
structed and tested for reliability. The equipment data was collected from the Ashagi Glass
Corporation (AGC) plant using the opinions of experts in the operations and maintenance
department. The nine factors which are most important and relevant are given weightings
and allocated overall scores from 1 to 9, which is 100%. In the questionnaire, each question
is assigned a five-point Likert score by assigning I–V weightings to maintain consistency of
degree: I – 20%; II – 40%; III – 60%; IV – 80%; V – 100%. To check whether the question-
naire measure the same latent variable of Cronbach’s alpha is run on a sample selected as
explained in the analysis section. The utmost critical issues are allotted the upper weightings
as related to less significant issues are shown in Table 3. Moreover, using the survey data all
of the equipment in the plant is categorized according to the criticality score, which is clas-
sified as follows: Critical – degree score above 80%; Very important – degree score between
60% and 80%; Important – degree score between 40% and 60%; Less important – degree
score below 40%.
Based on the above, a well-planned questionnaire was prepared and circulated to the heads
of important departments such as Operations, Maintenance, Inventory Control, Purchas-
ing, Quality Control, Fire and Safety, Planning, Electrical Maintenance, Instrumentation
and Control, Troubleshooting, Emergency Operations, Training, Computer Control Sec-
tion, Chemical Section, Hazard Control Section, and Research and Development Section to
gather their feedback. The results obtained are described in the following section.
4 RESULT ANALYSIS AND DISCUSSION
4.1 Reliability according to questionnaire

A group of 63 experts was selected from the total of 75. The chosen group included personnel
from service, production, and maintenance working in the department. Of the 63 experts, 59
replied to the questionnaire, which is shown in Appendix I. Only 56 responses were selected
as valid and were tested for internal consistency at a 95% confidence level using SPSS soft-
ware, which found that the Cronbach’s alpha was 0.82. The data represents presence of same
latent variable and found good. Respondents responded differently. The reliability statistics
are shown in Table 1.
Table 1. Reliability statistics.
Cronbach’s Cronbach’s alpha based No. of

alphaa on standardized items items
0.82 0.78 9
Table 2. Critical analysis output.
Total no. % of total no. Annual usage Critical

Class of items of items value (in lakhs) scores
Critical 62 26.38 9.5 80–100%

Important 121 51.48 6.9 40–80%
Less important 52 22.12 0.35 <40%
504

Figure 1. Trend in criticality of equipment.
Table 3. Criticality scores for equipment.
Degree I Degree II Degree III Degree IV Degree V

No Factor description Weighting (score) (score) (score) (score) (score)
1 % utilization of 10 M/c loaded 1 shift 2 shifts 3 shifts Overload

m/c for 8 hours (4) (6) (8) 10
(2)
2 External/internal 10 Very low Moderate Average High Extremely
pressure acting (2) (4) (6) (8) high
on m/c (10)
3 Availability of 10 4 or more 3 substitutes 2 substitutes 1 substitute No
substitute substitutes (4) (6) (8) substitute
(2) (10)
4 Impact on other 15 10% of 10–20% idle 20–40% idle 40–60% idle >80% idle
machines of machines or (6) (9) (12) (15)
breakdown men idle (3)
5 Age of m/c 10 Up to two years 2–4 years 4–6 years 6–8 years >8 years
(2) (4) (6) (8) (10)
6 Ease of repair 15 Mechanical Electrical and Electronic – –
fault easy to hydraulic fault
repair (5) fault difficult difficult
to diagnose to find
(10) (15)
7 Quality of work 5 Wide tolerance; Tolerance 0.05– Tolerance Tolerance Tolerance
done on m/c no effect on 0.10; no effect 0.025–0.05; 0.010–0.0250; 0.01;
quality on quality some effect major loss complete
(1) (2) on quality to production production
(3) (4) loss (5)
8 Ease of procurement 10 1 month; can be 1–3 months; 3–6 months; Over 6 months; To be
in terms of lead manufactured can be in-house in-house imported;
time/manufacture in the manufactured manufacture manufacture uncertain
in the company company in the not possible not possible procurement
(2) company (6) (8) period (10)
(4)
9 Maintenance 15 0–1 1–3 3–5 5–10 >10
history of m/c (3) (6) (9) (12) (15)
in terms of repair
orders received
Source: P & IC Hand Book.
4.2 Results of criticality analysis

Criticality analysis was carried out for the Ashagi Glass Corporation plant (a manufacturing
plant installed in a collaboration with Japan), at TCC Ltd. At first the criticality of machine
505

to which equipment belongs found the critical score of equipment. This was on the basis of
data collected from the plant with the help of the survey. They were classified as Critical,
Important or Less important. It was found that 62 items of equipment were Critical, 121
items were Important, and 52 items Less important, as shown in Table 2. A graph of the
trend is shown in Figure 1.
4.3 Discussion of criticality analysis

Damage of equipment in the Critical category will result in production stoppages or threats
to workers. Consequently, such equipment has never to be maintained freely and strategies
that will offer adequate safety must be implemented for these types. Important items must
be reasonably checked because their damage will have an effect on production for a shorter
period. Safe controls can be maintained for Less important items that will not create hazards
or impact production.
5 CONCLUSION
The identification of EIE is one of the key phases in elimination of accidents and severe
damage to production systems. The current investigation demonstrates that a methodical
approach has been implemented to distinguish the EIE and address problems arising in the
chemical processing industry. A number of proposals are suggested in our discussion of criti-
cality analysis for caring for such types of equipment and improving the existing maintenance
strategy. In addition, it is very clear from the literature survey that there are many methods
available for analyzing the reliability and maintenance program of equipment and lacks in to
identify the equipment with hazard involved in failure. This literature review is intended to
provide an idea of the preceding works that facilitated this work, helping to identify various
significant methodologies used in this field. Moreover, it assists in selecting the appropriate
procedure to identify the energy-intensified equipment for reliability analysis and the tool
for developing the maintenance program. The present work extends the scope of detailed
analysis in energy-intensified equipment.
REFERENCES
Barabady, J. & Kumar, U. (2008). Reliability analysis of mining equipment: A case study of a crushing
plant at Jajarm bauxite mine in Iran. Reliability Engineering and System Safety, 93, 647–653.
Ben-Daya, M. & Raouf, A. (1996). A revised failure mode and effect analysis model. International Jour-
nal of Quality & Reliability Management, 13(1), 43–47.
Eti, M.C., Ogaji, S.O.T. & Probert, S.D. (2006). Development and implementation of preventive-main-
tenance practices in Nigerian industries. Applied Energy, 83, 1163–1179.
Gilchrist, W. (1993). Modeling failure modes and effect analysis. International Journal of Quality and
Reliability Management, 10, 16–23.
Jayakumar, A. & Asgarpoor, S. (2004). Maintenance optimization of equipment by linear program-
ming. Probability in Engineering and Information Science, 20, 183–193.
Puthillath, B. & Sasikumar, R. (2012). Selection of maintenance strategy using failure mode effect and
criticality analysis. International Journal of Engineering and Innovative Technology, 1(6), 73–79.
Sachdeva, A., Kumar, D. & Kumar, P. (2009). Multi-factor failure mode critically analysis using TOP-
SIS. Journal of Industrial Engineering International, 5(8), 1–9.
Sanjeevy, C. & Thomas, C. (2014). Use and application of selective inventory control techniques of
spares for a chemical processing plant. International Journal of Engineering Research & Technology,
3(10), 301–306.
Tang, Y., Liu, Q., Jing, J., Yang, Y. & Zou, Z. (2017). A framework for identification of maintenance
significant items in reliability centered maintenance. Energy, 118, 1295–1303.
Teng, S.-H. & Ho, S.-Y. (1996). Failure mode and effects analysis: An integrated approach for
product design and process control. International Journal of Quality & Reliability Management,
13, 8–26.
506

APPENDIX I
TCC Limited, Ernakulam.

Questionnaire for Criticality Analysis
1. Machine/Equipment.............................................
2. Code.....................................................................
3. Specification........................................................
[Put tick mark ( ) for the appropriate options]
4. Percentage utilization of machine
a. M/c loaded for <8 hours b. 1 shift c. 2 shifts d. 3 shifts e. Overload
5. External/internal pressure acting on machine
a. Low b. Moderate c. Average d. High e. Extremely high
6. Availability of substitute
a. 4 or more substitutes b. 3 substitutes c. 2 substitutes d. 1 substitute e. No substitute
7. Impact on other machines of breakdown
a. 10% m/c or men idle b. 10–20% c. 20–40% d. 40–60% e. >80%
8. Age of machine
a. Up to two years b. 2–4 years c. 4–6 years d. 6–8 years e. >8 years
9. Ease of repair
a. Mechanical fault easy to repair
b. Electrical and hydraulic difficult to diagnose
c. Electronic fault difficult to find
10. Quality of work done on machine
a. Wide tolerance; no effect on quality
b. Tolerance 0.05–0.10; no effect on quality
c. Tolerance 0.025–0.05; some effect on quality
d. Tolerance 0.010–0.025; major loss to production
e. Tolerance <0.01; complete production loss
11. Ease of procurement in terms of lead time/manufacture in the company
a. 1 month/can be manufactured in the company
b. 1–3 months/can be manufactured in the company
c. 3–6 months/in-house manufacture not possible
d. Over 6 months/in-house manufacture not possible
e. To be imported; uncertain procurement period
12. Maintenance history of machine in terms of repair orders received
a. 0–1 b. 1–3 c. 3–5 d. 5–10 e. >10
507

International Conference on Advances
in Chemical Engineering (ICAChE)

Design of Fractional Filter Fractional Order Proportional Integral

Derivative (FFFOPID) controller for higher order systems
R. Ranganayakulu, G. Uday Bhaskar Babu & A. Seshagiri Rao

Department of Chemical Engineering, National Institute of Technology, Warangal, India
ABSTRACT: This article presents a Fractional Filter Fractional Order Proportional Inte-
gral Derivative (FFFOPID) controller design method for higher order systems, approximated
as Non-Integer Order Plus Time Delay (NIOPTD) models. The design uses an Internal
Model Control (IMC) scheme and the resulting controller has a series form of Fractional
Order Proportional Integral Derivative (FOPID) term, in series with a fractional filter. An
analytical tuning method is then used to identify the optimum controller settings. Simulation
results on different systems show that the proposed method gives better output perform-
ance for set point tracking, disturbance rejection, parameter variations, and for measurement
noise in the output. The robust stability of the system regarding process parametric uncer-
tainties is verified with robustness analysis. Controllability index analysis is also undertaken
to ascertain the closed loop system performance and robustness.
Keywords: Internal Model Control, robust stability, closed loop system, Fractional Order
Proportional Integral Derivative
1 INTRODUCTION
Higher order models describe the process dynamics more accurately than lower order models
(Isaksson & Graebe, 1999; Malwatkar et al., 2009). However, they complicate the controller
design and tuning for quality control. There are several controllers tuning rules for higher
order models, approximated as First Order Plus Time Delay (FOPTD) models. Most of these
rules are to tune a controller having a Proportional Integral Derivative (PID) structure, which
has been widely used to date (Aström & Hägglund, 1995; Skogestad, 2003). The controller
designed for such FOPTD models may not give the satisfactory performance, as the dynamics
are compromised during the approximation. An alternative to preserve the dynamics, while
ensuring satisfactory control, is to approximate them as Non-Integer Order Plus Time Delay
(NIOPTD) models (Pan & Das, 2013).The major advantage of NIOPTD models is that they
represent the process behavior more compactly than integer order systems (Podlubny, 1999).
Fractional order control for fractional order systems has been in focus in the last two
decades (Shah & Agashe, 2016). Several fractional order controller structures have been pro-
posed and the widely accepted one is the Fractional Order Proportional Integral Derivative
(FOPID) controller (Monje et al., 2008; Luo & Chen, 2009; Tavakoli-Kakhki & Haeri, 2011;
Padula & Visioli, 2011; Vinopraba et al., 2012; Das et al, 2011; Valério & da Costa, 2006).
The FOPID controller has the ability to enhance the closed loop performance, but the tun-
ing is complex as it has more tuning parameters than the PID controller. Recently, there was
work found in the literature where the five FOPID parameters are identified, based on the
stability regions of a closed loop system (Bongulwar & Patre, 2017). Further, the simulation
results were shown only for the servo response.
In this paper, a Fractional Filter Fractional Order Proportional Integral Derivative
(FFFOPID) controller is proposed using Internal Model Control (IMC). The present work
uses a series form of a FOPID controller (Hui-fang et al., 2015). The resulting controller has
511

a structure consisting of a FOPID term along with a fractional filter, with only two param-
eters to be tuned (Ranganayakulu et al., 2017). The tuning parameters are identified such
that the Integral Absolute Error (IAE) and Total Variation (TV) are minimal. The selection
is proved to be optimum by observing the trends of IAE and TV through an analysis, after
varying the tuning parameters in the range of +10% and −10%. The simulations have been
performed for different inputs and the measures used to assess the system performance are %
overshoot (%OS), Settling Time (ST), IAE and TV. Also, the applicability of the proposed
method for apparent changes in time delay is verified by varying the L/T ratio called as
controllability index (Ranganayakulu et al., 2016; Lin et al., 2008). The performance of the
proposed method is validated with three different examples.
This article is sectioned as follows: the IMC scheme is described in Section 2. Section 3
presents the controller design and tuning, and robust stability analysis is provided in Sec-
tion 4. The simulation results for different systems are presented in Section 5. The conclusion
is given in Section 6, and the paper ends with references.
2 PRELIMINARIES
2.1 Internal model control

The IMC method block diagram and feedback loop are shown in Figure 1. The IMC based
controller design procedure is briefly given below.
1. Identify the non-invertible (all time delays and unstable zeros) and invertible (minimum
phase elements) parts of the process model:
G ( s ) = G + ( s )G − ( s ) (1)
2. The IMC controller is:
f (s)
CIMC ( s ) = − (2)
G (s)
where f(s) is the IMC filter

3. The feedback controller is:
CIMC ( s )
C (s) = (3)
1− CIMC ( s )G ( s )
Figure 1. (a) IMC scheme; (b) feedback loop.
512

3 DESIGN OF FFFOPID CONTROLLER
The structure of the controller used here is:
 τ sλ + 1
C ( s ) = ( fractional filter )K p  i λ  ( 1 + τ d s µ ) (4)
 τ is 
To design the controller, the higher order system approximated as a NIOPTD model is
considered and is given by Equation 5:
Ke− Ls
G ( s ) = α (5)
Ts +1
where K, L, T and α are gain, time delay, time constant and fractional order of the model,
respectively. The feedback controller C(s), according to the procedure in Section 2.1 by
using the IMC filter f(s) = 1/(γsp+1) and first order Pade’s approximation for time delay
e-Ls = (1–0.5 Ls)/(1+0.5 Ls) is:

 1   L   1+ 0.5Ls 
C proposed ( s ) =   × 
 2 K   0.5Ls 
( 1+ Tsα ) (6)
 0.5γ Ls + γ s + L 
p p− 1
Comparing Equations 4 and 6, the settings are:
L 
Kp = ;τ i = 0.5L; λ = 1;τ d = T ; µ = α 
2K 
1  (7)
fractional filter = 
0.5γ Ls p + γ s p− 1 + L 
where γ and p are two tuning parameters of the controller.
3.1 Tuning
The tuning parameters γ and p are chosen in a way that the measures IAE and TV are minimal.
The optimum values are identified through the behavior of IAE and TV by varying γ and p in the
range of (-10%, +10%). Finally, γ is chosen for the minimum of both IAE and TV, and also p.
4 ROBUSTNESS ANALYSIS
The stability of a closed loop system should always be analyzed for process parameter uncer-
tainties because the process model is an approximation of the real plant. The robust stability
condition (Morari & Zafiriou, 1989) is:
lm ( s )T ( s ) < 1; s = jω ;∀ω ∈ (−∞ ,∞ ) (8)
where 1m ( s ) = G ( s ) − G ( s ) G ( s ) is the bound on multiplicative uncertainty and T(s) = C(s)

G(s)/1+C(s)G(s) is the complementary sensitivity function. For uncertainty in both L and K,
the following condition must be satisfied:
 ∆K  −∆L
T ( jω ) ∞ <1  +1 e − 1 (9)
 K 
Another constraint to be satisfied for robust control performance is:
513

T ( s )1m ( s ) +(1− T ( s )wm ( s )) <1 (10)
where 1-T(s) is the sensitivity function and wm(s) is the uncertainty bound on the sensitivity
function.
5 SIMULATION STUDY
Three higher order systems approximated as NIOPTD models are simulated in MATLAB and
the system performance is compared with the Bongulwar and Patre (2017) method (hereafter
addressed as the Patre (2017) method). The effectiveness of the proposed method is verified
with the performance measures %OS, ST, IAE and TV, which are defined in Table 1.Settling
time is defined as the time taken for the response to settle within 2% to 5% of its final value.
The closed loop system’s unit step response is observed, with step change in disturbance
of magnitude 0.5 applied at a later time. Also, the step response is observed for perturba-
tions of +10% in Land K and for measurement noise with a variance of 0.0001. The system
robustness for uncertainty is illustrated in the following sections through stability analysis.
The frequency used for Oustaloup approximation of fractional order is (0.01,100)rad/s. In
addition, the trend of closed loop behavior is interpreted for variation in controllability index
(i.e. L/T ratio in the range of 0.1 to 2). This analysis demonstrates the difficulty in control for
large changes in time delay.
5.1 Example 1
Consider the higher order system (Shen, 2002) and its equivalent NIOPTD (Patre, 2017)
model:
1 0.99149
G1 ( s ) = = e−1.6745 s (11)
( s +1)4 2.8015s1.0759 +1
The proposed and Patre (2017) controllers are:
 1   0.8372 s + 1  (12)
C proposed ( s ) =   × 0.99149   ( 1 + 2.8015s )
1.0759
 0.7535s1.01 + 0.9s 0.01 + 1.6745  0.8372 s
0.3432
Cold ( s ) = 1.5129 + + 0.1733s1.05 (13)
s1.1
The optimum values of γ and p for the proposed controller are identified as 0.9 and 1.01
(Figure 2). The performance measures for set point tracking are given in Table 2. The pro-
posed method is superior in performance compared to the Patre (2017) method with lower
values of %OS, ST, IAE and TV. The servo response for a disturbance applied at t = 30 s is
illustrated in Figure 3. Betterment is observed, even with disturbance with lower values of
performance measures, which is clear from Table 3. It is evident from Figure 3 that a satis-
factory performance is observed in terms of disturbance rejection. Figure 4 shows the step
response for a perturbed model, and the response for white noise in the measured output is
illustrated in Figure 5. It is observed that there is enhanced performance (Table 2) for both
Table 1. Expressions for %OS, IAE and TV.
%OS IAE TV
y peak − yss ∞ ∞
yss
× 100 ∫ e(t ) dt ∑ ui + 1 − ui
0 i= 0
514

Figure 2. Identification of optimum γ and p for example1.
Table 2. Servo performance comparison for the three different systems.
System Method %OS ST IAE TV
G1(s) Proposed 18.2 8.7 2.87 8.45

Patre (2017) 22.3 15 3.57 18.7
G2(s) Proposed 2.4 1.54 0.78 3.86
Patre (2017) 5.02 2.97 0.83 4.75
G3(s) Proposed 21.2 0.94 0.42 12.74
Patre (2017) 40.7 1.95 0.53 12.35
Figure 3. Closed loop response of G1(s) for step input.
Table 3. Comparison of IAE and TV for the three examples.
Perfect case Perturbed case Noise case
System Method IAE TV IAE TV IAE TV
G1(s) Proposed 3.984 9.2133 4.837 10.6533 4.281 38.7359

Patre (2017) 4.907 19.567 5.988 21.1318 5.02 137.5908
G2(s) Proposed 1.176 4.393 1.263 4.992 1.207 7.2042
Patre (2017) 1.243 5.3571 1.353 6.1191 1.248 10.4974
G3(s) Proposed 0.5663 13.5002 0.6612 16.185 0.6131 20.1658
Patre (2017) 0.6224 13.506 0.8196 20.2817 0.6767 20.6138
the cases with the proposed method, as compared to the Patre (2017) method. Also, there is
significantly less control effort with the proposed method for all the possible input changes.
The magnitude plot is shown in Figure 6 for +10% uncertainty in K; +10% and +50% uncer-
tainty in L. Robust stability condition in Equation 9 is violated by both the complementary
sensitivity functions for +50% uncertainty in time delay. The proposed method violates the
515

Figure 4. Closed loop step response of G1(s) for perturbations.
Figure 5. Step response in presence of measurement noise.
Figure 6. Magnitude plot for example1.
Figure 7. L/T ratio versus IAE, TV for step change in set point.
516

Figure 8. L/T ratio versus IAE, TV for step change in disturbance.
condition a bit earlier than the old method. Figure 7 and Figure 8 show the trends of IAE
and TV for servo and regulatory response with increasing L/T ratio. It is evident that increas-
ing trends are observed with the Patre (2017) method, compared to the proposed method.
Hence, the proposed method can be considered for enhanced closed loop performance of
processes with large changes in time delay.
5.2 Example 2
The second example (Chen et al., 2008; Patre, 2017) considered for performance comparison is:
9 1.0003
G2 ( s ) = = e−0.4274 s (14)
( s + 1)( s 2 + 2s + 9) 0.8864s1.0212 + 1
The proposed controller and the controller with Patre (2017) method are:

 1   0.2137 s + 1 
C proposed ( s ) = 
 0.07479s1.01 + 0.35s 0.01 + 0.4274 
× 0.2136 
 0.2137 s 
( 1+ 0.8864s1.0212 ) (15)
1.2203
Cold ( s ) = 1.4996 + + 0.0409s1.05 (16)
s1.05
The optimum values of γ and p are identified as 0.35 and 1.01. The closed loop system gives
good servo response with the proposed method. This is true with the lower values of %OS,
ST, IAE and TV given in Table 2. The unit step response with disturbance applied at t = 6 s
is shown in Figure 9 and the corresponding performance measures are given in Table 3. The
proposed method gives better servo response, which is evident with lower values of perform-
ance measures, while the regulatory performance is almost the same for both the control-
lers. The system response for perturbations is presented in Figure 10. Figure 11 presents the
closed loop response for white noise in the output. The proposed method continues to give
the superior performance compared to the Patre (2017) method, which is clear with the lower
values of IAE and TV (Table 3).
The closed loop robust stability for uncertainties in K and L is illustrated through the mag-
nitude plot in Figure 12. The closed loop system gives robust performance up to +100% uncer-
tainty in time delay and +10% uncertainty in gain with the proposed controller, whereas the
stability condition fails for +90% uncertainty in time delay with the Patre (2017) method. Fig-
ure 13 and Figure 14 show the trends of IAE and TV for servo and regulatory response with
increase in L/T ratio. The proposed method shows less control effort for servo and regulatory
response for the entire variation of L/T ratio. The trend followed by IAE for set point tracking
is almost the same up to L/T ratio of 1 for both the methods; after that, it starts increasing with
the old method. In the case of disturbance rejection, the IAE values are lower up to L/T ratio
of 1.3 with the old method, and then it increases. Hence, the proposed method is a good choice
to have a better control for increasing L/T ratio, compared to the old method (Patre, 2017).
517

Figure 12. Magnitude plot for example2.
518

5.3 Example 3
The higher order system studied in Panagopoulos et al. (2002) is considered as the third
example:

1 0.99932
G3 ( s ) = = e− . s (17)
( s + 1)(0.2 s + 1)(0.04 s + 1)( 0.008s + 1) 1.0842 s1.01332 + 1 0 1922
The proposed and old (Patre, 2017) controllers are given as follows:

 1   0.0961s + 1 
C proposed ( s ) =   × 0.0961   ( 1 + 1.0842 s ) (18)
1.0132
 0.0096 s1.1 + 0.1s 0.1 + 0.1922  0.0961s
6.14
Cold ( s ) = 5.0034 + + 0.0163s1.1 (19)
s1.1
The values of γ and p for the proposed method are 0.1 and 1.1. The performance measures
shown in Table 2 for servo response indicates that the %OS, ST and IAE values are lower
than with the proposed method, but that the TV value is slightly higher compared to the
old (Patre, 2017) method. Similarly, the step response for a change in disturbance applied at
t = 5 s is shown in Figure 15. The closed loop step response for process parameter variations
and for output noise is illustrated in Figure 16 and Figure 17. The corresponding perform-
ance measures for all the above cases are presented in Table 3. It is evident from all these
Figures that the proposed method gives superior servo performance but is a bit slow in reject-
ing the disturbance compared to the old method.
The proposed method gives robust performance up to an uncertainty of +70% in L and
+10% in K, while the Patre (2017) method fails for less than +50% uncertainty in L (Figure 18).
Figure19 and Figure 20 show the performance for variation of L/T ratio. For servo response,
the variation of IAE is low with the proposed method, while the control effort is slightly high
519

Figure 18. Magnitude plot for example 3.
520

up to L/T = 0.9, and then it increases drastically with the old method. In the case of regula-
tory control, the IAE values are higher with the proposed method and the variation of TV
is low. Hence, there is a trade-off between IAE and TV for increasing L/T, and the proposed
method is recommended for servo response, while it can be used for disturbance rejection at
higher values of L/T.
6 CONCLUSIONS
In this paper, a FFFOPID controller is proposed for higher order systems approximated as
NIOPTD models using IMC method. Analytical method is followed for identifying the tun-
ing parameters by minimizing IAE and TV. Enhanced closed loop performance is observed
with the proposed method for changes in set point and disturbance. In particular, the pro-
posed method is effective in terms of there being less control effort. The closed loop system is
robust with the proposed method for high uncertainty in process parameters. Also, the pro-
posed method assures better control for large changes in time delay, which is proved through
the variation of L/T ratio.
REFERENCES
Aström, K.J. & Hägglund, T. (1995). PID controllers: Theory, design, and tuning. Research Triangle
Park, NC: ISA.
Bongulwar, M.R. & Patre, B.M. (2017). Stability regions of closed loop system with one non-integer
plus time delay plant by fractional order PID controller. International Journal of Dynamics and Con-
trol, 5(1), 159–167.
521

Chen, Y.Q., Bhaskaran, T. & Xue, D. (2008). Practical tuning rule development for fractional order
proportional and integral controllers. ASME Journal of Computational Nonlinear Dynamics, 3(2),
021403.
Das, S., Saha, S., Das, S. & Gupta, A. (2011). On the selection of tuning methodology of FOPID con-
trollers for the control of higher order processes. ISA Transactions, 50(3), 376–388.
Hui-fang, W., Qiu-sheng, H., Zhi-cheng, Z & Jing-gang, Z.(2015). A design method of fractional order
PIλDµ controller for higher order systems. In 34th Chinese Control Conference (CCC), Hangzhou,
China (pp. 272–277).IEEE.
Isaksson, A. & Graebe, S. (1999). Analytical PID parameter expressions for higher order systems. Auto-
matica, 35(6), 1121–1130.
Lin, M.G., Lakshminarayanan, S. & Rangaiah, G.P. (2008). A comparative study of recent/popular PID
tuning rules for stable, first-order plus dead time, single-input single-output processes. Industrial &
Engineering Chemistry Research, 47(2), 344–368.
Luo, Y. & Chen, Y. (2009). Fractional order [proportional derivative] controller for a class of fractional
order systems. Automatica, 45(10), 2446–2450.
Malwatkar, M., Sonawane, S. & Waghmare, L. (2009). Tuning PID controllers for higher-order oscilla-
tory systems with improved performance. ISA Transactions, 48(3), 347–353.
Monje, C.A.,Vinagre, B.M., Feliu, V. & Chen, Y. (2008). Tuning and auto-tuning of fractional order
controllers for industry applications. Control Engineering Practice, 16(7), 798–812.
Morari, M. & Zafiriou, E. (1989). Robust process control. Englewood Cliffs, New Jersey: Prentice Hall.
Padula, F. & Visioli, A. (2011). Tuning rules for optimal PID and fractional order PID controllers. Jour-
nal of Process Control, 21(1), 69–81.
Pan, I. & Das, S. (2013). Model reduction of higher order systems in fractional order template. In Intel-
ligent Fractional Order Systems and Control (pp. 241–256). Berlin, Heidelberg: Springer.
Panagopoulos, H., Aström, K.J. & Hägglund, T. (2002). Design of PID controllers based on constrained
optimisation. IEEProceedings-Control Theory and Applications, 149(1), 32–40.
Podlubny, I. (1999). Fractional-order systems and PIλDµ controllers. IEEE Transactions on Automatic
Control, 44(1), 208–214.
Ranganayakulu, R., Babu, G.U.B., Rao, A.S. & Patle, D.S.(2016). A comparative study of fractional
order PIλ/PIλDµ tuning rules for stable first order plus time delay processes. Resource-Efficient Tech-
nologies, 2, S136–S152.
Ranganayakulu, R., Babu, G.U.B. & Rao, A.S. (2017). Fractional filter IMC-PID controller design for
second order plus time delay processes. Cogent Engineering, 4(1), 1366888.
Shah, P. & Agashe, S. (2016). Review of fractional PID controller. Mechatronics, 38, 29–41.
Shen, J.C. (2002). New tuning method for PID controller. ISA Transactions, 41(4), 473–484.
Skogestad, S. (2003). Simple analytic rules for model reduction and PID controller tuning. Journal of
Process Control, 13(4), 291–309.
Tavakoli-Kakhki, M. & Haeri, M. (2011). Fractional order model reduction approach based on reten-
tion of the dominant dynamics: Application in IMC based tuning of FOPI and FOPID controllers.
ISA Transactions, 50(3), 432–442.
Valério, D. & da Costa, J.S. (2006).Tuning of fractional PID controllers with Ziegler–Nichols-type rules.
Signal Processing, 86(10), 2771–2784.
Vinopraba, T., Sivakumaran, N., Narayanan, S. & Radhakrishnan, T.K. (2012). Design of internal
model control based fractional order PID controller. Journal of Control Theory and Applications,
10(3), 297–302.
522

Effect of parameters on electro-Fenton process for removal of oil

and grease from refinery wastewater
Mikhila Vikramaraj, A.M. Manilal & P.A. Soloman

Department of Chemical Engineering, Government Engineering College, Thrissur, Kerala, India
ABSTRACT: In the petroleum industry, oily water occurs in the stages of production, trans-
portation and refining, as well as during the use of derivatives. Crude oil is one of the major
components of wastewater from the petroleum industry. The project focuses on the use of the
Electro-Fenton (EF) technique for the removal of oil content from wastewater from the petro-
leum industry. Hydrogen peroxide is used as Fenton’s reagent and Fe2+ is provided from sacri-
ficial cast iron anodes. The crude oil samples are taken from the Cochin refinery. Experimental
study has been conducted to investigate the influence of various factors, for instance current
density, time of electro-Fenton, feed pH, and H2O2 concentration in the electro-Fenton process.
Keywords: Electro-Fenton technique, wastewater, Hydrogen peroxide
1 INTRODUCTION
Petroleum products play an unavoidable role in our daily lives and our demand is increasing day
by day. Petroleum products include transportation fuels, fuel oils for heating and electricity gener-
ation, and feed stocks for making the chemicals, plastics, and synthetic materials that are in nearly
everything we use. Of the approximate 7.19 billion barrels of total US petroleum consumption in
2016, 48% was motor gasoline (including ethanol), 20% was distillate fuel (heating oil and diesel
fuel), and 8% was jet fuel. So, these products have become essential in our routine lives.
These petroleum products are made from crude oil through a refining process. During the
refining process, petroleum refineries unavoidably generate large amounts of oily wastewater.
These wastes occur at different stages of oil processing, such as during production, transpor-
tation and refining. However, during the production phase large amounts of oily wastes are
generated, which become mixed with the sea water and cause pollution. Coelho et al. (2006)
reported that the quantity of water used in the oil refinery processing industry during the
production stage ranges from 0.4 to 1.6 times the volume of processed oil, and this wastewa-
ter may, if untreated, cause serious damage to the environment.
The presence of oil and grease in the water bodies accounts for a major part of water pol-
lution. Alade et al. (2011) explains the effects of these to the economy. The oil and grease will
form a thin oily layer above the water medium, which reduces the light penetration into the
water medium, thereby decreasing photosynthesis. Thus, it affects the survival of aquatic life
in water since the amount of dissolved oxygen in the water is less. It also affects the aerobic
and anaerobic wastewater treatment process due to the reduction in oxygen transfer rates,
and also due to the reduction in the transport of soluble substrates to the bacterial biomass.
So, more attention must be given to the treatment of oily wastewater. There are various
methods available for oil removal from wastewater, such as physical treatment, chemical
treatment, biological treatment, membrane treatment and advanced oxidation processing
(Krishnan et al., 2016).
The motivation of this project is the application of the Electro-Fenton (EF) process for
large-scale industries and wastewater treatment plants. Electro-Fenton treatment is regarded
as being a better mechanism for water treatment units.
523

In electro-Fenton, organic substances are removed at two stages of oxidation and coagula-
tion. Oxidation of organic substances is due to -OH radicals and coagulation is ascribed to
the formation of ferric hydroxo complexes (Nidheesh & Gandhimathi, 2012).
The degradation mechanism of organic pollutants by Fenton reaction is given in the fol-
lowing Equations.
Fe2+ + H2O2 → Fe3+ + OH- + HO •

(1)
RH + HO → R + H2O
• •
(2)
where, RH denotes organic pollutants
R + Fe3+ → R+ + Fe2+
•
(3)
Fe2+ + HO → Fe3+ + OH–

•
(4)
The viability of electro-Fenton is studied by conducting batch experiments for investigat-

ing the influencing parameters. These four parameters were identified viz. current density,
initial pH, hydrogen peroxide concentration, and reaction time.
The project mainly focuses on the study of different parameters that affect the electro-Fenton
process for removal of oil and grease content from crude oil wastewater. The electro-Fenton
process has two different configurations. In the first, Fenton reagents is added to the reac-
tor from outside and inert electrodes with high catalytic activity are used as anode material.
While in the second configuration, only hydrogen peroxide is added from outside and Fe2+
is provided from sacrificial cast iron anodes (Nidheesh & Gandhimathi, 2012). Here we are
using the second configuration. In this project we are using mild steel/iron as the anode and
stainless steel as the cathode. Therefore, an additional supplement of Fe2+ is not needed.

Figure 1 shows the experimental setup for the electro-Fenton process for the removal of oil
and grease from crude oil processed wastewater.
The reactor is designed for a volume of 2 liters of water to be treated. Therefore the dimen-
sions of the reactor selected are 25 cm × 10 cm × 10 cm, with a volume of 2.5 liters. The
reactor is made of plastic. The crude oil feed is fed into the reactor. The pH of the feed is
measured using a pH meter and the pH is adjusted by adding dilute HCl and NaOH. The
electro-Fenton process needs anodes and cathodes for the removal of crude oil. The anodes
are made of mild steel and the cathodes of stainless steel. The anodes and cathodes are
designed in such a way that they fit well in the reactor. The electrodes are placed in a parallel
arrangement. Here five anodes and four cathodes are used. The dimensions of the anodes
and cathodes are 8 × 5. The total area of ten electrodes was 4 dm2.
The hydrogen peroxide is added from outside as per required. Current and potential is
passed through the electrodes using a regulated DC power supply with the anode connected
to the positive, and the cathode to the negative ends of the DC power supply. The input volt-
age specification of the Aplab regulated power supply unit is 0–230V AC, and output voltage
specification is 0–30V DC. A magnetic stirrer was provided inside the reactor with varying
rotational speeds. The magnetic stirrer is an integral part of the mechanism because proper
contact of the electrodes with the wastewater is required to reduce the internal resistance
(IR) drop of the system. Once the experiment is completed, output samples are pippeted out
from the middle of the reactor. The samples are allowed to settle for half an hour and then
analyzed using Chemical Oxygen Demand (COD) and fluorescence spectrometry.
524

Figure 1. Electro-Fenton experimental setup.
Schematic of EF reactor – 1. Power supply with indication; 2. Reactor; 3. Magnetic stirrer; 4. Anodes
(mild steel); 5. Cathodes (stainless steel); 6. Water to be treated; 7. H2O2 added from outside.

The crude oil feed preparation procedure and electro-Fenton experimental procedure is
explained below.
2.2.1 Feed preparation

The crude oil from the desalter unit of Bharath Petroleum Corporation Limited (BPCL),
Kochi is taken for conducting the experiments. To prepare the feed, take 2,000 ml of water
from the reactor. Then 4 g crude oil, 1.5 g salt (NaCl) and 0.5 g SDS (Sodium Dodecyl
Sulfate) are added. Stir well using a magnetic stirrer at 500rpm for 10 mins. This is the feed.
The pH of the feed is measured using a pH meter. The pH is maintained according to experi-
mental requirement. It is maintained by adding dilute HSO4 or dilute NaOH. The SDS, a
surfactant, was added to help the oil particles better dissolve in the water. The characteristics
of feed are different for each set of experiments.
2.2.2 Experiment
Before moving to the experiment, first the electrodes are cleaned using dilute HCl. For that,
10 ml dilute HCl is used to clean the electrodes and the weights of the anodes and cathodes
are noted. Then we move on to the Fenton process. For that, 2 liters feed is taken from the
reactor. The electrodes (cathodes and anodes) are arranged bipolar. Electrodes are connected
to the DC supply. Current and voltage are adjusted. The experiment is run according to the
values of the parameters (current density, H2O2 concentration, pH, time) in the experiment
list. The salt concentration of the feed is 1.5 g.
After the experiment, the samples are collected and analyzed using COD value, fluores-
cence spectrometry and gravimetric analysis.
2.3 Experimental analysis

The experimental results are analyzed using the COD analyzer, fluorescence spectroscopy,
and gravimetric analysis.
2.3.1 Chemical oxygen demand

The samples are analyzed using COD. The COD value indicates the amount of oxygen needed for
the oxidation of all organic substances in water in mg/l. The COD is closely related to the labora-
tory standard dichromate-method. The method involves using a strong oxidizing chemical, potas-
sium dichromate, to oxidize the organic matter in solution to carbon dioxide and water under
525

acidic conditions. Often, the test also involves a silver compound to encourage oxidation of cer-
tain organic compounds, and mercury to reduce the interference from oxidation of chloride ions.
Based on this method, the COD has become a commonly used method in wastewater analysis.
2.3.2 Fluorescence spectroscopy

The Hitachi Fluorescence Spectrophotometer Model F-4600 is also used for the analysis of
the samples. It is a research-grade fluorescence spectrophotometer with high sensitivity and
high scanning speeds, capable of multi-mode analysis.
In this project fluorescence analysis for the oil in water was conducted. Fluorescence is
the emission of light by a substance that has absorbed light. In general, the emitted light
has a longer wavelength, and therefore lower energy, than the absorbed radiation. A light
source is provided inside the instrument. A monochromatic filter is provided for selecting the
excitation wavelength. Then the spectral distribution of the light emitted from the sample is
analyzed. This is detected by a light detector. The sample was placed inside the fluorescence
spectrometer in a quartz cuvette for the analysis.
The project focuses on removal of oil and grease from crude oil wastewater by the electro-
Fenton process. COD and analysis of oil and grease using a fluorescence spectroscope and
gravimetric analysis are used to estimate the percentage oil removal in each experiment. The
samples from each experiment are analyzed using COD and fluorescence spectroscope and
gravimetric method and the percentage removal of oil and grease is found by comparing with
the feed values. The COD value gives the amount of organic substances found in the water.
Therefore, it is an important measure of water quality.
3.1 Calibration of fluorescence spectroscopy

Fluorescence spectroscopy is used to find the oil content from the fluorescence values of
different emission wavelengths which were supplied with a constant excitation wavelength.
The working parameters for conventional spectra of crude oil were an excitation wave-
length of 380 nm, an emission range of 200–900 nm (5 nm intervals) with 5.0 nm excitation,
and 5.0 nm emission slits with a scan rate of 1,200 nm/min (Sotelo et al., 2008). Fluores-
cence values were calibrated in terms of concentration in ppm. The calibration was done with
known concentrations. The calibration curve is shown in Figure 2.
From the calibration curve:
Fluorescence = 0.0788 × Concentration + 3.2801
From this the concentration was calculated as:
Concentration (ppm) = 12.488 (Fluorescence – 37.765)
The excitation wavelength of crude oil is given as 380 nm. The fluorescence spectrometer
generates the spectrum. The emission wavelength is given in the range of 200–900 nm. The
height at emission wavelength apex 418.2 nm is taken for noting the crude oil concentration.
3.2 Effect of parameters on electro-Fenton process

The preliminary stage experiments were conducted to analyze the effect of different param-
eters that influence the electro-Fenton process. The individual influence of a parameter was
studied by keeping all other parameters constant.
3.2.1 Influence of pH
The pH value will affect the oxidation and coagulation of the electro-Fenton process. The
impact of initial pH value on percentage removal was studied. The initial pH was adjusted
using 0.1 N HCl and 0.1 N NaOH. The initial pH value was found using a digital pH meter.
526

Figure 2. Calibration curve for fluorescence spectroscopy – fluorescence v/s concentration.
The experiment was conducted at current density 0.5 A/dm2, reaction time of 20 mins, and
H2O2 concentration 0.275 g/l.
The initial pH plays an important role in the electro-Fenton process. Figure 3 shows the percent-
age of oil removal using COD and fluorescence analysis. The study was conducted at a pH range
from 2–7. Generally, the Fenton process was conducted at a pH below 7 (i.e. at acidic medium).
From the results it was clear that the maximum removal was obtained at a pH around
4. The removal becomes less effective at a pH < 3. It is due to regeneration of Fe2+ ions,
through reaction between Fe3+ and H2O2. At higher pH the removal also decreases rapidly:
at a pH > 5, the % removal was found to decrease to a value of 50%. It is due to the fact that
H2O2 is unstable in basic solution (Nidheesh & Gandhimathi, 2012).
3.2.2 Influence of current density

Current density is the electric current per unit area of cross section of the anode. As the cur-
rent increases, current density also increases.
The maximum current density taken was 0.75 A/dm2 and the oil removal was found to
increase with an increase in current density. The experiment was conducted at pH 4, reaction
time 20 mins and H2O2 concentration 0.275 g/l. As current density increases, removal per-
centage also increases. The maximum removal is obtained at maximum current density (i.e.
at a current density of 0.73 A/dm2). The applied current is the driving force for the reduction
of oxygen, leading to the generation of hydrogen peroxide at the cathode.
3.2.3 Influence of H2O2 concentration

Hydrogen peroxide is used as Fenton’s reagent, which plays a major role in the removal of oil.
The removal efficiency increases with the increase in hydrogen peroxide concentration. For
the experimental analysis, the amount of hydrogen peroxide taken for a 2 liter crude oil feed
is 0.42 g, 0.51 g, 0.55 g, 0.6 g, and 0.68. All other parameters were kept constant (i.e. current
density 0.5 A/dm2, reaction time 20 mins, and at pH 4).
Figure 5 shows that as the amount of hydrogen peroxide increases, initially the removal
increases and then decreases. The maximum efficiency was obtained at 0.55 g hydrogen
peroxide for 2 liters of crude oil feed. As the H2O2 concentration increases, the increase in
removal efficiency is due to the increase in hydroxyl radical concentration. At high dosage
of H2O2, the decrease in removal efficiency was due to the scavenging effect of H2O2 and the
recombination of hydroxyl radical (Yu et al., 2013).
3.2.4 Influence of reaction time
The last parameter analyzed was reaction time. Time increment resulted in a sharp increment
in the removal efficiency.
This might be due to the Farady’s law of electrolysis, which gives a direct relationship
between time and the amount of anode material released. A saturation point was observed at
around 20 mins for the electro-Fenton process. There was not much increase in removal after
the saturation value, as can be seen in Figure 6.
527

Figure 3. Variation of % oil removal with Figure 4. Variation of % oil removal with
change in pH. change in current density.
Figure 6. Variation of % oil removal with

Figure 5. Variation of % oil removal with change change in time.
in H2O2 concentration.
4 CONCLUSION
The work focused on the removal of oil and grease from crude oil processed wastewater
releasing from refineries. Crude oil wastewater from BPCL, Kochi was selected as the sample
for treatment purpose. For the removal of oil and grease from crude oil processed wastewater,
the electro-Fenton technique was used. Both oxidation and coagulation contributed to COD
removal through Fenton treatment of wastewater.
COD test and fluorescence spectrometry analysis were the major analyses conducted for
the treated water. Electro-Fenton experiments were conducted with mild steel and stainless
steel electrodes. The important parameters affecting the EF process were analyzed, –viz. pH,
current density, H2O2 concentration and time. At pH 4 percentage removal was maximum,
and decreased with an increase and decrease in the pH of water. The oil removal increases
with increase in current density. At highest current density, the maximum removal efficiency
was obtained. In the case of H2O2 concentration, the maximum removal was obtained at
0.55 g/l with a reaction time of 30 mins.
REFERENCES
Alade, A.O., Jameel, A.T., Muyibi, S.A., Karim, M.I.A., & Alam, Z. (2011). Application of semifluidized
bed bioreactor as novel bioreactor system for the treatment of palm oil mill effluent (POME). Afri-
can Journal of Biotechnology, 10(81), 18642–18648.
528

Al-Harbawi, A.F.Q., Mohammed, M.H. & Yakoob, N.A. (2013). Use of Fenton’s reagent for removal
of organics from Ibn Al-Atheer Hospital wastewater in Mosul City. Al-Rafidain Engineering Journal,
21(5), 127–135.
Atmaca, E. (2009). Treatment of landfill leachate by using electro-Fenton method. Elsevier Journal of
Hazardous Materials, 163(1), 109–114.
Coelho, A., Castro, A.V. Dezotti, M. & Sant’Anna Jr, G.L. (2006). Treatment of petroleum refinery
sourwater by advanced oxidation processes. Elsevier Journal of Hazardous Materials, 137(1), 178–184.
Gordon Jr, D.C., Keizer, P.D. & Dale, J. (1974). Estimates using fluorescence spectroscopy of the present
state of petroleum hydrocarbon contamination in the water column of the northwest Atlantic Ocean.
Elsevier Scientific Publishing Company. Amsterdam. Marine Chemistry, 2(4), 251–261.
Jameel, A.T., Muyubi, S.A., Karim, M.I.A & Alam, M.Z. (2011). Removal of oil and grease as emerging
pollutants of concern (EPC) in wastewater stream. IIUM Engineering Journal, 12(4), 161–169.
Kalra, S.S., Mohan, S., Sinha, A. & Singh, G. (2011). Advanced oxidation processes for treatment of
textile and dye wastewater: A review. In 2nd International Conference on Environmental Science and
Development IPCBEE, 4, 271–275.
Krishnan, S., Chandran, K., & Sinnathambi, C.M. (2016). Wastewater treatment technologies used for
the removal of different surfactants: A comparative review. International Journal of Applied Chemis-
try, 12(4), 727–739.
Nidheesh, P.V. & Gandhimathi, R. (2012). Trends in electro-Fenton process for water and wastewater
treatment: An overview. Elsevier Desalination, 299, 1–15.
Patil, A.D. & Raut, P.D. (2014). Treatment of textile wastewater by Fenton’s process as an advanced
oxidation process. IOSR Journal of Environmental Science, Toxicology and Food Technology (IOSR-
JESTFT). e-ISSN: 2319–2402. p-ISSN: 2319-2399, 8, 29–32.
Soloman, P.A., Basha, C.A., Velan, M. & Balasubramanian, N. (2009). Electrochemical degradation
of pulp and paper industry waste-water. Journal of Chemical Technology and Biotechnology, 84(9),
1303–1313.
Sotelo, F.F., Pantoja, P.A., López-Gejo, J., Le Roux, G.A.C., Quina, F.H. & Nascimento, C.A.O. (2008).
Application of fluorescence spectroscopy for spectral discrimination of crude oil samples. Brazilian
Journal of Petroleum and Gas, 2(2), 63–71.
Yu, R.F., Lin, C.H., Chen, H.W., Cheng, W.P. & Kao, M.C. (2013). Possible control approaches of the
electro-Fenton process for textile wastewater treatment using on-line monitoring of DO and ORP.
Elsevier Chemical Engineering Journal, 218, 341–349.
Zhang, J. & Dong, H. (2009). Study of treatment of oil pollution from water with electro-Fenton tech-
nology. IEEE. 978-1-4244-2902-8/09.
529

Conversion studies of methanol to olefin on boric

acid treated Al/MCM-41
D. Kumar, N. Anand & A. Kedia

University School of Chemical Technology, GGS IP University, Delhi, India
ABSTRACT: The use of a large-pore catalyst with a high surface area, MCM-41, loaded
with alumina, is explored for the production of ethylene and propylene from methanol con-
version. MCM-41 is a silica-based catalyst with low Lewis acid site strength. Al/MCM-41
was treated with boric acid at three different concentrations. A modified catalyst was char-
acterized using BET, chemisorptions, XRD and SEM. The total surface area was observed
to reduce after boric acid treatment with the treatment of MCM-41. The maximum decrease
in surface area was obtained with the treatment of Al/MCM-41 with boric acid (1M). An
N2 adsorption-desorption plot shows a change in the porous structure of the catalyst after
treatment with boric acid. The conversion studies were performed at different temperatures
between 250–450°C and liquid flow rate in the range of 30 to 120 ml/min. The effect of the
catalyst on the selectivity of ethylene and propylene was studied with Al/MCM-41 and B-Al/
MCM-41. Results showed that the boric acid treated Al/MCM-41 helps to increase the selec-
tivity of the catalyst toward propylene production. Gas yield was also observed to increase
after using the boric acid treated catalyst. A 20.3% decrease in the coke yield was observed
when the experiments were performed with the boric acid treated catalyst as compared to an
untreated catalyst. However, no significant effect on the coke production was obtained in the
presence of three boric acid treated Al/MCM-41 catalysts.
Keywords: large-pore catalyst, boric acid, propylene production, coke production
1 INTRODUCTION
In the recent past, the scientific community has been putting more effort into looking for
alternative routes for the production of feedstock materials required in the petrochemical
industries. At present, olefins are mainly produced from methanol conversion. The main
source of methanol is methane, obtained from the petroleum refineries (Tian et al., 2015).
However, as petroleum refineries are now being modified to process the gas obtained from
the natural resources, new and effective technologies are needed to convert methane to meth-
anol and further feedstock materials such as DME and formaldehyde. In coming years, more
methane will be obtained from shale gas or gas hydrates (Lefevere et al., 2014).
Zeolites such as ZSM-5 has been widely used throughout the world, due to its shape selec-
tivity, durability and reusability for a wide range of reactions in the petroleum refinery and
petrochemical industries (Khare et al., 2017). However, the small size of cage and small pore
size of ZSM-5 are its major drawbacks in the Methanol to Olefin (MTO) process, wherein
large-size products cannot escape from the small cage opening, thereby deactivating the cata-
lyst. To overcome the fast deactivation issue of catalysts, a large-pore silica-based mesopo-
rous catalyst has been used for different reactions at various laboratories in the last decade,
and is a subject of interest (Wu et al., 2012). These catalysts have both two-dimensional and
three-dimensional structures with large-pore diameters and high surface area (e.g. SBA-15,
MCM-41, MCM-22, FDU-13 and MFU). However, these catalysts lack mechanical strength
and have low acid site concentrations required for reactions (Olsbye et al., 2012; Li et al.,
531

2011; Almutairi et al., 2013; Xu et al., 2013). These catalysts can be tailored for their pore
size as well as acid site concentration. The surface functionalization of the hydroxyl groups
with alumina has affected the acidity of these silicious catalysts. Studies on alumina and
other metals like Pd, Ga, Cu, and Ge impregnated MCM-41, have been reported. These
metals help in modifying the Lewis as well as Bronsted acid sites, for various reactions on
mesoporous as well as microporous catalyst, including ZSM-5, for various reactions (Linares
et al., 2014; Naik et al., 2010; Bhattacharyya et al., 2003; Sang et al., 2013). As mentioned by
a few authors, faster deactivation of catalyst might occur due to loading of alumina. But is
reported that the loading of alumina might be helpful in increasing the mechanical strength
and increasing the number of strong acid sites. The modification of the acid site concentra-
tion by the treating of ZSM-5 with boric acid, or oxalic acid, or phosphoric acid have been
reported (Du et al., 2006; Murthy et al., 2010; Sang et al., 2013; Epelde et al., 2015). This
modification might be applied for the conversion of methanol to gasoline hydrocarbons.
This work reports on the modification of alumina-loaded MCM-41 with boric acid to
study the effect of change in the type of acid site concentration, and thus the conversion of
methanol and selectivity for ethylene and propylene in a fixed-bed quartz reactor.
2.1 Catalyst modification

Alumina-loaded MCM-41 was purchased from ACS Material LLC and was modified with
boric acid. The solid powder was mixed in 0.5, 1.5 and 2.5M H3BO3 with solution/catalyst
ratio of 20 cm3/g at 80°C for 8h, and calcined at 550°C for 6h.
2.2 Catalyst characterization

The purchased catalyst and the modified catalyst were characterized for their properties,
such as surface area, surface structure, crystalline structure and mesoporous structure. Fou-
rier Transform Infrared Spectroscopic (FT-IR) studies of the catalysts were obtained on a
Nicolet Model: 6700 in the wavelength range from 4,000 to 500 cm–1. The catalysts, before
and after boric acid treatment, were first dried at 110°C, and then 0.1 g of each catalyst
was mixed with KBr and pressed in a pelletizer to form pellets of 1 mm thickness. X-ray
diffraction spectra were obtained on a Philips X’Pert diffractometer, equipped with a mono-
chromatic CuKα, with radiation having an angular range (2θ) from 0.5 to 5°; preliminary
phase analysis was performed with a scan speed of 1 s per step and a step size of 0.05°. N2
adsorption-desorption studies were performed on a Quantachrome Autosorb iQ instrument
at 77°K. Both the catalysts were out-gassed in vacuum for 3h at 300°C. Both the mesoporous
and the overall surface area were obtained with the help of using the BET method. NH3
temperature-programmed desorption (NH3-TPD) was performed with a temperature-pro-
grammed desorption system. The ammonia desorption amount was measured as a function
of temperature. The distribution of type of hydrocarbons in the liquid product was measured
using a PerkinElmer GC/Mass Spectrometer (GC/MS) Clarus 500, fitted with an Elite-5MS
column (30 m length, 0.25 mm i.d. and 0.5 µm film thickness) with helium as the carrier gas.
The oven program was started at 35°C for 8 minutes, 100°C for 5 minutes, 150°C for 2 min-
utes, and 250°C for 40 minutes. The mass spectra were obtained at the ionization energy of
70 eV from m/z 12/400.

The performance studies of the purchased and modified catalysts were performed in a quartz
reactor (2 cm diameter and 40 cm length). In a typical run 0.5 g of catalyst was packed in the
middle of reactor, with the help of glass beads and glass wool packed at the two ends, with
a N2 flow at 50 mL/min. The studies were conducted at a temperature range of 250–450°C
532

and a liquid flow rate of 30–120 ml/min. Methanol with water (in a ratio of 1:4) was made to
first enter a pre-heater fixed at 150°C with the help of a syringe pump, and the vapors were
then made to pass through the catalyst bed. The vapors from the reactor were then passed via
a condenser maintained at 4°C, with the help of a chiller using water and isopropyl alcohol.
The condensed liquid and uncondensed vapors from the condenser were further sent to a
separator, in which liquid was collected at different time intervals and gases were injected into
the GC for the analysis. The conversion and ethylene/propylene selectivity were evaluated as
follows:
FMethanol, fed − FMethanol, out

X Methanol (%) = (1)
FMethanol, fed
yield of ith product

Si (%) = (2)
total gas products
2.4 Product analysis

Uncondensed gases obtained from the outlet of the separator were injected into the GC
TCD, as well as the FID equipped with a Porapak column. The estimation of the gases
such as carbon dioxide, carbon monoxide, methane, ethane, ethylene, propane, propylene,
butane, butane, butylenes, pentane and pentene was performed, and then its quantification
was obtained using standards available. The analysis of liquid products was performed on
the GC/MS for estimation of distribution of the components, and on the GC equipped with
capillary column for the quantification estimation of conversion of methanol and yield of
other compounds such as DME.
3.1 Characterization of catalyst

FT-IR spectra of both the treated and untreated catalysts are shown in Figure 1. As shown
in the Figure, the intense band at 1,050 cm–1 confirms asymmetric stretching of the Al-O-Al
or Si-O-Si groups, whereas the peak at 810 cm–1 confirms the symmetric stretching of Al-O-
Al or Si-O-Si groups. The band at 955 cm–1 shows the presence of defective Si-OH groups,
and the band at 1,640 cm–1 is the bending vibration absorption of OH (Bhattacharyya et al.,
2003). These bands confirm that Al/MCM-41 and B-Al/MCM have a skeleton similar to that
of MCM-41.
The XRD of both the catalysts Al/MCM-41 and B-Al/MCM-41 were performed and the
peaks were obtained at different theta values of 2.30, 3.94 and 4.54, resulting from the (100),
(110) and (200) crystal faces of MCM-41. These results confirmed that even after alumina
loading and boric acid treatment, the ordered hexagonal-lattice mesopore structure is sus-
tained (Du et al., 2006). However, a decrease in the intensity of the diffraction peak at (100)
and (200) was observed after boric acid treatment. The N2 adsorption-desorption isotherms
and the pore size distributions obtained from chemisorption for treated and untreated Al/
MCM-41 is shown in Figure 2.
As shown in Figure 1, all the N2 adsorption-desorption isotherms were observed to be
similar to those of type IV, indicative of the existence of mesopores. BET analysis showed
that the Al-MCM-41 has higher surface areas (890 m2/g) with pore diameter of 3.15, which
was observed to decrease to 782, 693 and 601 m2/g on treatment with boric acid 0.5, 1.5
and 2.5M, respectively. However, no significant decrease in the pore diameter was observed
after boric acid treatment. For studying variations in the surface acidity and strength of acid
sites, NH3-TPD characterization was used for the Al/MCM-41 before and after boric acid
treatment. The peaks at 210 and 330°C were obtained for weak acidic sites, whereas the peak
533

Figure 1. FT-IR spectra of the catalyst.
Figure 2. N2 adsorption-desorption curves of Al/
MCM-41 and boric acid treated Al-MCM-41.
obtained at 500°C indicates the strong acidic sites. In general, Al impregnated MCM-41 has a
higher strength of strong acid site, which was observed to decrease after boric acid treatment.
A slight enhancement was observed after boric acid treatment. Based on the strength of
weak acid site concentration and strong site concentration, Al/MCM-41 treated with boric
acid (1.5 M) was selected for further studies.
3.2 Catalyst performance studies

3.2.1 Methanol conversion and gaseous product yield at different temperature
Conversion of methanol was performed at 250, 300, 350, 400 and 450°C in the fixed-bed
quartz reactor for Al/MCM-41 and B-Al/MCM-41 (1.5M) and is shown in Figure 3. Lower
conversions, 56.3% and 67.1%, were obtained at 250°C. However, as the temperature was
increased from 250–450°C, above 90% conversions were obtained for Al/MCM-41 and B-Al/
MCM-41 (1.5M), respectively. Higher change (approximately 12%) in the yield of liquid
products was observed (from 19.2% to 7.2%) with increase of temperature from 250°C to
450°C in the presence of boric acid treated Al/MCM-41, whereas the gas yield was increased
from 78.3% to 91.4%. A similar pattern was observed with Al/MCM-41; however, the per-
centage changes were different.
3.2.2 Methanol conversion and gaseous product yield at different flow rate
The conversion studies were performed with 0.5 g catalyst but at a flow rate range from
30 ml/hr to 120 ml/min, keeping the nitrogen flow rate at 50 ml/min and a methanol-to-
water ratio of 1:2 (Figure 4). The maximum methanol conversion (89.2%) was obtained
at 90 ml/hr with B-Al/MCM-41, whereas with Al/MCM-41 catalyst the maximum conver-
sion was 83.4%. A similar pattern in gaseous products yield was observed (90.3%), whereas
maximum gas yield with Al/MCM-41 was 85.3%. Lower liquid yield was obtained in both
the cases.
3.2.3 Selectivity of ethylene and propylene

The gases coming from the outlet of the reactor were analyzed and found to include methane,
ethylene, propylene, butane and pentane in the presence of Al/MCM-41 and B-Al/MCM-41.
Table 1 shows the various conversion and selectivity of various hydrocarbons. The selectiv-
ity of ethylene was 4.3% with Al/MCM-41 and 7.1% with B-Al/MCM-41, indicating a 7.2%
decrease. However, the selectivity for propylene was increased from 8.1% to 15.3%. No major
change was observed in the selectivity of C4 hydrocarbons between 7.3% and 10.4%. C5 was
approximately 10.2% and was observed to decrease with boric acid treatment. It has been
534

Figure 3. Methanol conversion at different Figure 4. Methanol conversion at different
temperatures. temperatures.
Table 1. Catalytic performance for MTO reaction.
Selectivity (%)
Catalyst Conv (%) C1–4 C2H4 C3H6 C4H8 C5+
Al/MCM-41 92.1 13.2 4.3 8.1 7.3 10

B-Al/MCM-41 (1.5M) 94.1 6.3 7.1 15.3 10.4 3.7
mentioned by a few reporters that a moderate decrease in strong acid site presence due to
decrease in alumina sites is responsible for increase in selectivity for propylene.
4 CONCLUSIONS
The conversion of methanol for the production of ethylene and propylene was performed
in a fixed-bed reactor. The highest conversions were obtained at 450°C and at a flow rate of
90 ml/min. Alumina-loaded MCM-41 was treated with boric acid and the selectivity of pro-
pylene was observed to increase from 8.1% to 15.3%. The selectivity of ethylene was observed
to decrease and was mainly observed to promote the propylene yield.
REFERENCES
Abrokwah, R.Y., Deshmane, V.G. & Kuila, D. (2016). Comparative performance of M-MCM-41 (M:
Cu, Co, Ni, Pd, Zn and Sn) catalysts for steam reforming of methanol. Journal of Molecular Cataly-
sis A: Chemical, 425, 10–20.
Almutairi, S.M.T., Mezari, B., Pidko, E.A., Magusin, P.C.M. & Hensen, E.J.M. (2013). Influence of
steaming on the acidity and the methanol conversion reaction of HZSM-5 zeolite. Journal of Cataly-
sis, 307, 194–203.
Bhattacharyya, K.G., Talukdar, A.K., Das, P. & Sivasanker, S. (2003). Al-MCM-41 catalysed alkylation
of phenol with methanol. Journal of Molecular Catalysis A: Chemical, 197(1–2), 255–262.
Du, G., Lim, S., Yang, Y., Wang, C., Pfefferle, L. & Haller, G.L. (2006). Catalytic performance of vana-
dium incorporated MCM-41 catalysts for the partial oxidation of methane to formaldehyde. Applied
Catalysis A: General, 302(1), 48–61.
Epelde, E., Santos, J.I., Florian, P., Aguayo, A.T., Gayubo, A.G., Bilbao, J. & Castañoa, P. (2015). Con-
trolling coke deactivation and cracking selectivity of MFI zeolite by H3PO4 or KOH modification.
Applied Catalysis A: General, 505, 105–115.
535

Khare, R., Liu, Z., Han, H. & Bhan, A. (2017). A mechanistic basis for the effect of aluminum content
on ethene selectivity in methanol-to-hydrocarbons conversion on HZSM-5. Journal of Catalysis, 348,
300–305.
Lefevere, J., Mullens, S., Meynen, V. & Van Noyen, J. (2014). Structured catalysts for methanol-to-
olefins conversion: A review. Chemical Papers, 68(9), 1143–1153.
Li, J., Wei, Y., Liu, G., Qi, Y., Tian, P., Li, B., … Liu, Z. (2011). Comparative study of MTO conversion
over SAPO-34, H-ZSM-5 and H-ZSM-22: Correlating catalytic performance and reaction mecha-
nism to zeolite topology. Catalysis Today, 171(1), 221–228.
Linares, N., Silvestre-Albero, A.M., Serrano, E., Silvestre-Albero, J. & Garcíıa-Martínez, J. (2014). Mes-
oporous materials for clean energy technologies. Chemical Society Reviews, 43(22), 7681–7717.
Murthy, K.V.V.S.B.S.R., Kulkarni, S.J., Chandrakala, M., Mohan, K.K., Pal, P. & Rao, T.P. (2010).
Alkylation of 1-naphthol with methanol over modified Silicoaluminophosphate and MCM-41 molec-
ular sieves. Journal of Porous Materials, 17(2), 185–196.
Naik, S.P., Bui, V., Ryu, T., Miller, J.D. & Zmierczak, W. (2010). Al-MCM-41 as methanol dehydration
catalyst. Applied Catalysis A: General, 381(1–2), 183–190.
Olsbye, U., Svelle, S., Bjorgen, M., Beato, P., Janssens, T.V.W., Joensen, F., & Lillerud, K.P. (2012). Con-
version of methanol to hydrocarbons: How zeolite cavity and pore size controls product selectivity.
Angewandte Chemie International Edition, 51(24), 5810–5831.
Sang, Y., Li, H., Zhu, M., Ma, K., Jiao, Q. & Wu, Q. (2013). Catalytic performance of metal ion
doped MCM-41 for methanol dehydration to dimethyl ether. Journal of Porous Materials, 20(6),
1509–1518.
Sang, Y., Liu, H., He, S., Li, H., Jiao, Q., Wu, Q. & Sun, K. (2013). Catalytic performance of hierarchi-
cal H-ZSM-5/MCM-41 for methanol dehydration to dimethyl ether. Journal of Energy Chemistry,
22(5), 769–777.
Tian, P., Wei, Y., Ye, M. & Liu, Z. (2015). Methanol to olefins (MTO): From fundamentals to commer-
cialization. American Chemical Society Catalysis, 5(3), 1922–1938.
Wu, L., Degirmenci, V., Magusin, P.C.M.M., Szyja, B.M. & Hensen, E.J.M. (2012). Dual template
synthesis of a highly mesoporous SSZ-13 zeolite with improved stability in the methanol-to-olefins
reaction. Chemical Communications, 48(76), 9492–9494.
Xu, A., Ma, H., Zhang, H., Ying, W. & Fang, D. (2013). Conversion of methanol to propylene over a
high silica B-HZSM-5 catalyst. International Journal of Chemical, Molecular, Nuclear, Materials and
Metallurgical Engineering, 7(4), 175–184.
536

Degradation of diphenamid by UV/hydrogen peroxide advanced

oxidation process
M.S. Manju & V. Nishan Ahammed

K.P. Prasanth Kumar

Department of Chemistry, Maharaja’s College, Ernakulam, India
ABSTRACT: Degradation and mineralization of diphenamid by hydrogen peroxide in the

presence of UV was investigated. The results indicate that a diphenamid sample treated with
hydrogen peroxide along with UV irradiation has the maximum degradation in the given
time. Hydrogen peroxide alone can give considerable degradation, but hydrogen peroxide
together with UV can maximize the extent as well as providing a better scheme of degrada-
tion. The study indicates that UV irradiation with the injection of 80 microliters of 20 v/v
hydrogen peroxide to 1 liter of 0.1 ppm sample solution can completely disintegrate the pes-
ticide content in two hours.
Keywords: diphenamid, UV, hydrogen peroxide, Degradation, mineralization
1 INTRODUCTION
Recently, Advanced Oxidation Processes (AOPs) have emerged as promising methods for the
removal of organic pollutants from water. AOPs are based on the use of hydroxyl radicals
for oxidative disintegration of organic pollutants into environmentally benign substances
such as CO2 and H2O. Diphenamid (DPA) is a herbicide used for controlling annual grasses
and weeds in tomato, potato, peanut, and soybean plants (Schultz & Tweedy, 1972; Sirons
et al., 1981). As in the case of other pesticides and herbicides, DPA also enters into the water
bodies and poses a threat to the environment in general, and to the aquatic organisms in par-
ticular. Therefore, the development of effective methods for remediation of polluted water
containing even trace amounts of DPA is significant. Researchers have established that toxic
pollutants impact on the health of the ecosystem and present a threat to humans through the
contamination of drinking water supplies (Eriksson et al., 2007).
Several researchers had studied the photochemical degradation of DPA in aqueous solu-
tion. Rosen (1967) studied the homogeneous photodegradation of DPA by UV and sun-
light irradiation. Rahman et al. (2003) investigated the photocatalytic degradation of DPA in
aqueous P25 TiO2 suspension under the illumination of a medium-pressure mercury lamp.
Liang et al. (2010) studied the homogeneous and heterogeneous degradations of DPA in
aqueous solution by direct photolysis with UVC (254 nm) and by photocatalysis with TiO2/
UVA (350 nm). H2O2-based AOP studies on the degradation of diphenamid have not been
reported by researchers so far.
The objective of the present work is to study the application of UV/H2O2-based AOP for
the removal of DPA from water. Degradation of the pesticide is not the only concern for us,
but the compounds and the intermediates which are formed during the course of these reac-
tions are also of utmost importance. Hence, we have employed the most advanced analytical
537

techniques in this work, viz. the application of High-Performance Liquid Chromatography
(HPLC) and a Total Organic Carbon (TOC) analyzer. Designing a method with a safe deg-
radation pathway is always a challenging task. It is equally important to develop an eco-
nomically viable and time-bound method which easily blends in with the existing purification
methods. This makes this work exciting and socially relevant.
2.1 Materials
The diphenamid we used was 99.99% pure, purchased from Sigma Aldrich.
The hydrogen peroxide solution was standard lab quality, which is 20 v/v.
2.2 Methods
2.2.1 Extent of degradation analysis
2.2.1.1 Preparation of standard solution for extent of degradation analysis
DPA solution of 1,000 ppm (1 g of diphenamid dissolved in 1,000 ml solution) was magneti-
cally stirred for 60 min. All reactions were performed at room temperature.
2.2.1.2 TOC analysis of sample with UV degradation
About 200 ml of the DPA standard solution was taken in a beaker and placed in a UV reac-
tor. The solution was thoroughly stirred using a magnetic stirrer. The reactor started, and the
samples were collected from the solution at regular intervals of time.
2.2.1.3 TOC analysis of sample with H2O2 degradation
About 200 ml of the DPA standard solution was taken in a beaker, placed on a magnetic stir-
rer and thoroughly stirred. To this solution 10 microliters of 20 v/v H2O2 was added and the
initial sample was collected. Then the timer was started, and the samples were collected from
the solution at regular intervals.
2.2.1.4 TOC analysis of sample with H2O2 and UV degradation
About 200 ml of the DPA standard solution was taken in a beaker and placed in a UV reac-
tor. The solution was thoroughly stirred using a magnetic stirrer. To this solution 10 microlit-
ers of 20 v/v H2O2 was added and the initial sample was collected. The timer was started, and
the samples were collected from the solution at regular intervals.
2.2.1.5 Analytical method
The samples obtained are analyzed using a TOC analyzer. Initially the analyzer was calibrated
using a blank sample. After that, each sample was analyzed using a suitable method. Here,
the method used was the Non-Purgeable Organic Carbon (NPOC) measurement method.
2.2.2 Effect of H2O2 loading

2.2.2.1 Preparation of standard solution for degradation time analysis
Prior to photolytic reaction, the DPA solution of 0.1 ppm (0.1 mg of diphenamid dissolved
in 1,000 ml solution) was magnetically stirred for 60 min. All reactions were performed at
room temperature.
2.2.2.2 Methodology for reaction with UV irradiation: Standard sample without H2O2
About 250 ml of the DPA standard solution was pipetted into a beaker and placed in a UV
reactor. The samples were collected from the solution at different intervals. The DPA concen-
tration was determined using HPLC.
2.2.2.3 Methodology for reaction with UV irradiation: Sample with H2O2
About 250 ml of the DPA standard solution was pipetted into a beaker and placed in a UV
reactor. To this solution 10 microliters of 20 v/v H2O2 was added and the initial sample was
collected. Samples were collected from the solution at different intervals. The DPA concentra-
tion was determined using HPLC. Then the procedure was repeated for 20 microliters of H2O2.
538

2.2.2.4 Analytical method
Samples obtained were analyzed using HPLC. Initially HPLC was calibrated using a standard
solution. Thus, retention time was obtained along with the peak area for different samples.
Later, each sample was analyzed. The peak areas obtained were consolidated and standard-
ized to the scale from 0 to 1 and plotted.
3.1 Extent of degradation

3.1.1 Result sample 1 – Diphenamid sample irradiated with UV
The reaction of diphenamid was conducted in the immersion type photo reactor as five-hour
long runs, resulting in the following observations:
i. Degradation rate in the first hour was very low and only 0.1% of the total diphenamid
present in the solution was degraded.
ii. The degradation rate in the second hour was comparatively faster, but there was only 0.5%
of degradation.
iii. The degradation rate in the third hour was the fastest in the entire run, and 1.2% of the
total diphenamid present in the sample was degraded.
iv. The degradation pattern in the fourth hour seems similar to the degradation pattern in
the second hour. This period showed a degradation of 0.4% and was considered to be the
slowest rate in the entire run.
v. The degradation pattern in the fifth hour seemed similar to the degradation pattern in the
third hour, with a degradation of 1%.
vi. In the five hours, UV irradiation had given a total degradation of 3.1% of the total
diphenamid present in the solution.
Figure 1 shows the fraction of diphenamid in the sample vs time plot. Overall, degrada-
tion rate in this sample run is very low. The degradation pattern is non-uniform, showing a
wide variation in the degradation rate. Irradiation caused only a total degradation of 3.1%
in five hours.
3.1.2 Result sample 2 – Diphenamid sample mixed with hydrogen peroxide

The experimental run of diphenamid sample mixed with hydrogen peroxide was conducted
for five hours as per the standard procedure. The observations are noted below:
i. Degradation rate exhibited a drastic increase to 18% in the first hour. This was the maxi-
mum degradation rate in the entire run.
ii. The degradation rate in the second hour seemed to be very low. In this period the degrada-
tion was only 0.95%.
Figure 2. TOC analysis data of sample

Figure 1. TOC analysis data of sample treated with hydrogen peroxide.
irradiated with UV.
539

iii. The degradation rate in the third hour was lower compared to the degradation in the sec-
ond hour. In this period the degradation was only 0.29%.
iv. As time passed, the degradation pattern showed a reduction in the rate, as in the fourth
hour the degradation was further reduced to 0.11%. In the fifth hour the degradation was
as low as 0.06%.
v. In five hours, there was a total degradation of 23% of the diphenamid present in the
solution.
Figure 2 shows the fraction of diphenamid in the sample vs time plot. The degradation
rate was very high in the beginning, but quickly reduced with time. The degradation rate was
nearly zero in the fifth hour. The degradation pattern shows that the initial hike in the deg-
radation rate may be attributed to the fast and spontaneous reaction due to the addition of
hydrogen peroxide, which is reduced with increase in time. The five hours of stirring after the
addition of hydrogen peroxide brought about around 23% degradation of the diphenamid
initially present in the solution.
3.1.3 Result sample 3 – Diphenamid sample treated with H2O2 irradiated by UV

The photo reaction of diphenamid solution mixed with hydrogen peroxide was conducted
as a continuous run for five hours, as per the standard procedure. The observations are as
given below:
i. The degradation in the first hour seemed to be moderate with degradation of 13% of the
initial concentration of the diphenamid in the solution.
ii. The degradation rate was lower in the second hour. In this period the extent of degrada-
tion was 6.5%.
iii. In the third hour the degradation pattern showed a further reduction behavior with a
degradation of 3.1%.
iv. The reduction in degradation with respect to time was observed in the fourth hour too,
with a degradation of 1.5%.
v. However, in the fifth hour there was a variation with a slight increase in the degradation
rate, giving a degradation of 1.7%.
vi. In five hours, UV irradiation had given a total degradation of 26.8% of the total diphena-
mid present in the solution.
Figure 3 shows the fraction of diphenamid in the sample vs time plot. The degradation
pattern shows a uniform reduction of degradation rate with time. Initial hours of reaction
show the maximum degradation rate with 13% degradation. However, degradation rates in
the fourth to fifth hours show an increase. The sample exhibits a non-uniform degradation
pattern with alternate increase and decrease. Overall degradation was 26.8% in five hours.
Figure 3. TOC analysis data of sample Figure 4. Cn/C0 data plot for extent of degra-
treated with hydrogen peroxide/UV. dation analysis.
540

A diphenamid sample with added hydrogen peroxide and irradiated by UV has a degrada-
tion pattern which moreover shows a uniform reduction of degradation rate with time. Initial
hours of reaction showed the maximum degradation rate with 13% degradation. However,
degradation seems to be active after five hours with an increase in the degradation rate.
Also, the pattern exhibits a non-uniform degradation rate reduction behavior with alternative
increase and decrease. An overall degradation of 26.8% happened in five hours.
Figure 4 shows a comparison plot of degradation rate of all the above samples. It is clear
that UV irradiation gave the least extent of degradation in the given time, showing that UV
is incapable for total degradation of the compound. It can be only used for sterilization in
order to eliminate microorganisms. A diphenamid sample treated with hydrogen peroxide
shows a high degradation of 18% in the beginning, but the total degradation percentage
was limited to 23%, while a diphenamid sample with hydrogen peroxide and irradiated/UV
treatment shows comparatively less degradation of 13% initially, but made a total degrada-
tion of 26.8%.
The observations from the extent of degradation analysis confirm that a diphenamid sam-
ple with added hydrogen peroxide and irradiated by UV has the maximum degradation in
the prescribed time of five hours. The addition of hydrogen peroxide can give considerable
degradation, but hydrogen peroxide together with UV can give the maximum degradation as
well as a better degradation scheme, since hydrogen peroxide can degrade pesticides and UV
can destroy microorganisms.
3.2 Effect of amount of H2O2 on extent of degradation

Figure 5 shows the effect of H2O2 loading on the rate of degradation of diphenamid. The
degradation curve for diphenamid solution UV irradiated without hydrogen peroxide shows
a very slow degradation pattern. Degradation rate continuously decreases with time. About
95% of the total diphenamid in the solution degrades in the first four hours. However, total
degradation time was found to be five hours.
When the diphenamid solution is UV irradiated after adding 10 microliters of hydrogen
peroxide, the sample shows a fast degradation pattern. The degradation rate is very high in
the first hour, degrading 79% of the diphenamid present. The curve shows a stagnation rate
in-between one to two hours. The total degradation time is around 4 hours.
The degradation curve for the diphenamid solution UV irradiated after adding 20 micro-
liters of hydrogen peroxide shows a drastic degradation pattern, with 91% of diphenamid in
the sample degraded in one hour. The degradation rate shows a decreasing pattern with time
in general. The total degradation time is around two hours.
The results show that UV irradiation followed by the addition of hydrogen peroxide leads
to a much faster degradation of diphenamid. Also, degradation time decreases with increase
in the volume of hydrogen peroxide. Only 20 microliters of hydrogen peroxide was sufficient
to degrade 250 ml of 0.1 ppm diphenamid solution in two hours. That means that 80 ml of
Figure 5. Cn/C0 data plot for effect of H2O2 loading.
541

20 v/v hydrogen peroxide is sufficient to degrade 1,000l of water with 0.1 ppm diphenamid
concentration.
4 CONCLUSIONS
The results of the present study confirm that a diphenamid sample with added hydrogen per-
oxide and irradiated by UV has the maximum degradation in the prescribed time of five hours.
The addition of hydrogen peroxide can give considerable degradation, but hydrogen peroxide
together with UV can give maximum degradation as well as a better degradation scheme,
since hydrogen peroxide can degrade pesticides and UV can destroy microorganisms.
However, the knowledge about the compounds formed due to the degradation of diphena-
mid and the behavior and properties of the intermediate compounds in the reaction pathway
are beyond the scope of this work, which will definitely serve as a scope for future investiga-
tions in this topic.
ACKNOWLEDGMENT
One of the authors (Manju M.S.) acknowledges with thanks the financial assistance received
from Centre for Engineering Research (CERD), Government of Kerala.
REFERENCES
Eriksson, E., Baun, A., Mikkelsen, P.S. & Ledin, A. (2007). Risk assessment of xenobiotics in stormwa-
ter discharged to Harrestup Å, Denmark. Desalination, 215(1–3), 187–197.
Liang, H.C., Li, X.Z., Yang, Y.H. & Sze, K.H. (2010). Comparison of the degradations of diphenamid
by homogeneous photolysis and heterogeneous photocatalysis in aqueous solution. Chemosphere,
80(4), 366–374.
Rahman, M.A., Muneer, M. & Hahnemann, D. (2003). Photocatalysed degradation of a herbicide
derivative, diphenamid in aqueous suspension of titanium dioxide. Journal of Advanced Oxidation
Technologies, 6(1), 100–108.
Rosen, J.D. (1967). The photolysis of diphenamid. Bulletin of Environmental Contamination and Toxi-
cology, 2(6), 349–354.
Schultz, D.P. & Tweedy, B.G. (1972). Effect of light and humidity on absorption and degradation of
diphenamid in tomatoes. Journal of Agriculture and Food Chemistry, 20(1), 10–13.
Sirons, G.J., Zilkey, B.F., Frank, R. & Paik, N.J. (1981). Residues of diphenamid and its phytotoxic
metabolite in flue-cured tobacco. Journal of Agriculture and Food Chemistry, 29(3), 661–664.
542

Assessment of trihalomethanes in drinking water using

gas chromatography
S.P. Aravind & P.A. Soloman

Department of Chemical Engineering, Government Engineering College Thrissur, Kerala, India
ABSTRACT: Water is an essential element to sustain life. To ensure this requires a safe,
adequate and accessible supply. Therefore, efforts should be made to achieve a standard
drinking water quality. This is achieved through the use of water treatment plants, where
the major objective is the removal of pathogenic microorganisms to prevent the spread of
water-borne diseases. It is important that water treatment works be equipped with adequate
disinfection systems. Disinfection processes can result in the formation of both organic and
inorganic Disinfection By-Products (DBPs). The most well-known of these are the organo-
chlorine by-products such as Trihalomethane (THM) compounds. THM concentrations in
drinking water are measured using Gas Chromatography (GC) at various places in Thrissur
City, including at the Government Engineering College, which has its water treated at the
Peechi water treatment plant. A study has been conducted on parameters that affect
the formation of THMs. Response surface designs have been created for each THM using the
design of an experiment tool in Minitab 17, with the most influencing parameters being
observed from parameter study.
Keywords: Disinfection, Trihalomethane, THM, RSM
1 INTRODUCTION
There are many sources of contamination in drinking water, ranging from natural substances
leaching from soil to harmful chemical discharges from industrial plants. In developing coun-
tries, nearly half of the population is suffering due to lack of potable water, or due to con-
taminated water (WHO, 1992). Since disinfection is the most popular step in the treatment of
water, the cheapest method is preferred. Hence, chlorination is the most common disinfection
method as it remains in water until it has been consumed (Sadiq & Rodriguez, 1999). How-
ever, this poses a chemical threat to human health as it reacts with organic matter available in
the water to produce harmful products.
During the chlorination of water containing organic matter, different Disinfection By-Prod-
ucts (DBPs) are formed and more than 300 different varieties have been identified (Becher,
1999). In the range of 37–58% of the total measured halogenated by-products are trihalometh-
anes. Trihalomethanes (THMs) are a group of four volatile compounds that are formed when
chlorine reacts with organic matter present in the water (Frimmel & Jahnel, 2003).
The THMs include: Trichloromethane (Chloroform-CF), Bromodichloromethane
(BDCM), Dibromochloromethane (DBCM) and Tribromomethane (Bromoform-BF).
These are classified as possible human carcinogens by the US Environment Protection Act
(USEPA) (US Environmental Protection Agency, 1990). In the USEPA guidelines, the maxi-
mum contaminant level specified in the DBP Stage I Rule is 80 µg/L for total THMs (US
Environmental Protection Agency, 1998). The formation of these compounds depends on
several other factors such as temperature, pH, disinfectant dose, contact time, inorganic
compounds and organic matter present in the drinking water supply (Bull et al., 1995; Wu
et al., 2001; Bach et al., 2015).
543

In India, surface water, which can be divided into river, lake, and complex types, provides
more than 90% of the country’s drinking water. Accordingly, as the water sources are differ-
ent, their characteristics differ and eventually the amount of THMs also differs. Therefore,
we measured THM forming according to the quality of the raw water and the possible fac-
tors favoring the formation.
The objective of this study was to analyze trihalomethanes present in the available drink-
ing water using Gas Chromatography (GC), and to study the parameters that influence the
formation of THMs, to find optimum values and to model them.
2.1 Chemicals
• Certified reference material of THM 2,000 µg mL–1 in methanol was purchased from
Supelco, USA.
• n-hexane of HPLC grade (99%) was purchased from Sigma-Aldrich.
• Hypochlorite solution (13%) was purchased from suppliers.
• Working standards were prepared using n-hexane.
• Other chemicals (e.g. NaOH, H2SO4, NH4Cl) were analytical grade.
2.2 Extraction and analysis

The standalone method for the extraction is liquid-liquid extraction, and for analysis is gas
chromatography with electron capture detector (GC-ECD) or gas chromatography mass
spectrometry (Frimmel & Jahnel, 2003; Siddique et al., 2015; Kim et al., 2002). Here THMs
are extracted with n-hexane and analysis done using a Perkin Elmer Clarus 580 gas chro-
matograph, equipped with a headspace injection system and an electron capture detector
(9 in. × 9 in. × 9.8 in. = 794 in3. Its maximum usable depth = 6.3 in. It accepts 1/8 in. o.d.
stainless steel, 6 mm o.d. glass and all fused silica, packed or capillary columns 6.5 in.
diameter coil).
The method was developed by tuning the GC, based on the data from the supplier of the
certified reference material and from the handbook of water treatment technologies (WHO,
2008; The Environmental Protection Agency, 1998). The tuned GC operating parameters for
the analysis are given in Table 1. The GC is calibrated using five working standards.
2.3 Sample analysis

Water samples were collected three times at two-week intervals from various parts of Thris-
sur City to determine the concentration of THMs. Seven of the samples were from water
treated at the Peechi water treatment plant, and two were from the pump houses at the Gov-
ernment Engineering College, Thrissur. Water from well 1 of the college was used as raw
water for the parameter study.
Table 1. Tuned GC operating parameters for THM analysis.
Parameter Conditions
Oven Temperature 80°C for 4 minutes

Injector Temperature 150°C
Detector Temperature 300°C
Carrier Gas Flow Nitrogen gas (99.998%) @ 30 mL/min
Injection Volume 1 μL
544

2.4 Experimental set up for parameter study
Water samples from well 1 at Government Engineering College, Thrissur were collected for
the parameter study. 100 mL of sample water was taken in a 500 mL beaker, which was then
placed over a temperature-programmed magnetic stirrer. Experiments were carried out by
varying one parameter but keeping the other parameters constant. This scheme was repeated
for all parameters. Parameters chosen were chemical oxygen demand, HOCl concentration,
temperature, pH, reaction time and mixing time.
2.5 Modeling using Minitab 17

There were 27 sets of experiments suggested by the response surface methodology tool of
Minitab 17 for four sets of parameters that have more influence on THM formation. A
response surface design refers to a set of advanced Design of Experiments (DOE) tech-
niques, which help in experimental analysis and response optimization. Box-Behnken design
was selected for the design of experiments.
The response Minitab optimization tool helps to identify the combination of input vari-
able settings that jointly optimize a single response or a set of responses. Minitab calculates
an optimal solution and draws a plot.
3.1 Determination of THMs in samples

THM concentrations of the samples were analyzed three times. Their average concentrations
in ppb are given in Table 2. The observed concentrations are far below permissible limits.
3.2 Parameter study

The concentrations of each THM present in the raw water is given in Table 2, and the lev-
els seem to be very low. Chlorination experiments were carried out and the influence of all
parameters in the formation of THMs were studied. It is observed that THM concentra-
tion increases with HOCl concentration, reaction time and temperature. However, the nature
fluctuated for pH and mixing time. Also, THM increases to a particular level for COD, after
which it decreases, except in the case of chloroform.
3.3 Response surface design

Four parameters were selected that have more influence in THM formation, as is seen from
parameter study. Parameters and their domains are given in Table 3.
Table 2. Results of water sample analysis.
Location CF BDCM DBCM BF
Swaraj Round 3.5574 2.1885 0.6818 0.0208

Arimboor
0.8597 0.0341 0.0101 0.0025
Panchayath
Koorkenchery 10.1506 4.0765 1.6565 0.1391
Ramavarmapuram 9.5372 3.5249 0.2425 0.0168
Peringavu 0.9445 0.6641 1.3561 0.0480
Poonkunnam 5.0526 3.3114 1.2292 0.0971
Pump House 1,
0.1004 0.0081 0.0003 0.0000
GEC Thrissur
Pump House 2,
3.3620 2.9177 0.0019 0.0215
GEC Thrissur
Raw Water 0.1004 0.0081 0.0003 0
545

Table 3. The level and range of variables.
Domain
Parameters Factor -1 1
COD (ppm) A 40 800

HOCl Concentration (ppm) B 0.2 3
Temperature (°C) C 20 50
Reaction Time (min) D 30 90
Table 4. Response surface design data.
Run COD HOCl Temperature Reaction CF BDCM DBCM BF

order (ppm) (ppm) (°C) time (min) (ppb) (ppb) (ppb) (ppb)
1 40 1.6 35 90 0.9915 0.1036 0.02610 0.006880

2 420 1.6 35 60 0.3701 0.0101 0.00220 0.001000
3 420 3.0 20 60 0.3770 0.1502 0.01790 0.000000
4 420 1.6 20 90 0.3230 0.0219 0.00580 0.002200
5 420 1.6 35 60 0.3701 0.0101 0.00220 0.001000
6 40 1.6 35 30 0.3727 0.0349 0.00720 0.004700
7 800 1.6 35 90 0.4557 0.0071 0.00350 0.001700
8 420 0.2 50 60 0.1197 0.0004 0.00110 0.000100
9 420 0.2 20 60 0.3684 0.0035 0.00000 0.002700
10 40 1.6 50 60 0.4482 0.0527 0.01850 0.006500
11 420 1.6 50 30 0.4464 0.0668 0.01280 0.006400
12 420 1.6 50 90 0.1971 0.0189 0.00440 0.005500
13 800 1.6 50 60 0.2357 0.0221 0.00300 0.001500
14 420 3.0 35 90 0.5922 0.1611 0.02650 0.007332
15 40 1.6 20 60 0.3665 0.0591 0.01630 0.004490
16 420 0.2 35 90 0.2404 0.0072 0.00320 0.001700
17 800 0.2 35 60 0.4394 0.0038 0.00370 0.002200
18 800 3.0 35 60 0.5700 0.0305 0.00160 0.000400
19 420 3.0 35 30 0.3051 0.0548 0.00480 0.002700
20 420 3.0 50 60 0.4553 0.1231 0.01860 0.008800
21 420 1.6 20 30 0.2232 0.0736 0.01250 0.005200
22 40 0.2 35 60 0.3850 0.0001 0.00100 0.000300
23 800 1.6 35 30 0.8377 0.1234 0.01862 0.007700
24 800 1.6 20 60 0.2812 0.0193 0.00090 0.005490
25 40 3.0 35 60 0.4790 0.0924 0.02430 0.008800
26 420 1.6 35 60 0.3701 0.0101 0.00220 0.001000
27 420 0.2 35 30 0.2909 0.0038 0.00140 0.003200
The three level second-order designs demand comparatively less experimental data to ena-
ble precise prediction. In the Box-Behnken method a total number of 27 experiments, includ-
ing three center points, are carried out to estimate the formation of THM. The quality of the
fit of this model is expressed by the coefficient of determination R2. The concentration of
each THM after the chlorination experiment is subtracted from its level before chlorination,
and the concentrations are given in Table 4.
Models have been created to determine the concentration or quantity of each compound
based on the data. Regression equations for each component in uncoded units are given by:
a. Chloroform
CF formed = −0.435 + 0.000427 a −0.201 b + 0.0388 c + 0.00419 d + 0.000001 a*a −0.000480
c*c + 0.000079 d*d −0.000022 a*d + 0.00389 b*c + 0.00201 b*d −0.000194 c*d.
546

b. Bromodichloromethane
BDCM formed = 0.182 + 0.000192 a −0.0250 b −0.00646 c −0.00313 d + 0.000000 a*a + 0.01142
b*b + 0.000089 c*c + 0.000031 d*d −0.000031 a*b −0.000004 a*d + 0.000612 b*d.
c. Dibromochloromethane
DBCM formed = 0.0226 + 0.000031 a −0.00001 b −0.000786 c −0.000513 d + 0.000000
a*a + 0.00095 b*b + 0.000013 c*c + 0.000006 d*d −0.000012 a*c −0.000000 a*c −0.000001
a*d −0.000005 b*c + 0.000118 b*d −0.000001 c*d.
d. Bromoform
BF formed = 0.01870 + 0.000015 a −0.00385 b −0.000526 c −0.000302 d + 0.000000
a*a + 0.000007 c*c + 0.000003 d*d −0.000005 a*b −0.000000 a*c −0.000000 a*d + 0.000136
b*c + 0.000037 b*d.
where a is COD, b is HOCl concentration, c is temperature and d is reaction time. R2 values
obtained are 80.78%, 78.64%, 84.15%, and 89.46% respectively.
Response optimization has given COD 646.46 ppm, HOCl concentration 0.20 ppm, tem-
perature 50°C, and reaction time 90 minutes, for obtaining minimum concentration values
of all THMs. The minimum values obtained for BF, DBCM, BDCM and CF are −0.0011,
−0.006, −0.0216 and −0.0968 respectively. Figure 1 shows the results given by the response
optimizer tool of RSM for obtaining minimal formation of THMs.
Since the RSM optimized the system to a possible minimum combination, slightly nega-
tive values are obtained. In order to find the combination where zero THM are formed, the
optimization is performed using the Microsoft Excel 2016 Solver tool.
The combinations observed are given in Table 5. Thus, the combination of parameters
were obtained for zero production of CF, BDCM, DBCM and a near to zero value for BF.
Figure 2 shows some of the contour plots of different THMs obtained using the Min-
itab 17 tool. It gives the trend of compound formation by considering two different varying
parameters and two hold parameters.
Figure 1. Results of response optimizer for THM.
Table 5. Optimized combinations given by solver.
a b c d THM Conc.
Analytes (ppm) (ppm) (°C) (min) (ppb)
CF 438.53 0.2 50 84.85 0

BDCM 800 0.2 34.55 45.76 0
DBCM 556.21 3 50 90 0
BF 40 0.2 35.63 49.10 0.0024
547

Figure 2. Contour plots of THMs.
4 CONCLUSION
Disinfection is a crucial and necessary step in the drinking water treatment process. Trihalom-
ethanes are the major disinfection by-products formed in drinking water when organic matter
present in the water reacts with chlorine. It is very toxic and carcinogenic and causes severe health
effects. THM concentrations are determined from water samples taken at two-week intervals
from various parts of Thrissur City. The deviations from mean concentration values are not at all
high. The concentrations of all the components are below that of the highest permissible level.
COD, HOCl concentration, temperature and reaction time are observed to be more influ-
encing parameters from the parameter study. Response surface designs were created using
Minitab 17 for all components. By knowing the parameters’ values, uncoded coefficients and
regression equation, it is possible to calculate the formation of individual THM components
with around 83% reliability. The Minitab response optimizer tool gave optimized values of:
COD 646.46 ppm, HOCl concentration 0.20 ppm, temperature 50°C, and reaction time
90 minutes, for the minimal formation of THMs. The Microsoft Excel solver tool optimized
the values of parameters to achieve zero THM.
REFERENCES
Bach, L., Garbelini, E.R., Stets, S., Peralta-Zamora, P. & Emmel, A. (2015). Experimental design as a
tool for studying trihalomethanes formation parameters during water chlorination. Microchemical
Journal, 123, 252–258.
Becher, G. (1999). Drinking water chlorination and health. CLEAN–Soil, Air, Water, 27(2), 100–102.
Bull, R.J., Birnbaun, L.S., Cantor, K.P., Rose, J.B., Butterworth, B.E., Pegram, R. & Tuomisto, J. (1995).
Water chlorination: Essential process or cancer hazard? Toxicological Sciences, 28(2), 155–166.
Frimmel, F.H. & Jahnel, J.B. (2003). Formation of haloforms in drinking water. In Haloforms and
Related Compounds in Drinking Water, 5(Part G) (pp. 1–19). Springer, Berlin: Heidelberg.
Kim, J., Chung, Y., Shin, D., Kim, M., Lee, Y., Lim, Y. & Lee, D. (2002). Chlorination by-products in
surface water treatment process. Desalination, 151(1), 1–9.
Sadiq, R., & Rodriguez, M.J. (2004). Disinfection by-products (DBPs) in drinking water and predictive
models for their occurrence: A review. Science of the Total Environment, 321(1–3), 21–46.
Siddique, A, Saied, S., Mumtaz, M., Hussain, M.M. & Khwaja, H.A. (2015). Multipathways human
health risk assessment of trihalomethane exposure through drinking water. Ecotoxicology and Envi-
ronmental Safety, 116, 129–136.
The Environmental Protection Agency (1998). Water Treatment Manual: Disinfection.
US Environmental Protection Agency. (1990). Risk assessment, management and communication of
drinking water contamination, EPA/600/4-90/020. Washington, DC.
US Environmental Protection Agency. (1998). National primary drinking water regulations; disinfectants
and disinfection by-products; final rule, fed. regist., 63(241), 69389–69476.
World Health Organization. (1992). Our planet our health: report of the WHO commission health and
environment. Geneva: World Health Organization.
World Health Organization. (2008). Guidelines for drinking-water quality [electronic resource]: 1st and
2nd addenda, vol. 1, Recommendations.
Wu, W.W., Benjamin, M.M., & Korshin, G.V. (2001). Effects of thermal treatment on halogenated dis-
infection by-products in drinking water. Water Research, 35(15), 3545–3550.
548

Adsorption of perchlorate using cationic modified rice husk
Ann M. George
Department of Chemical Engineering, University of Kerala, Kerala, India
K.B. Radhakrishnan
Department of Chemical Engineering, TKM Engineering College, Kollam, India
A. Jayakumaran Nair
Department of Biotechnology, Kariavattom Campus, Thiruvananthapuram, India
ABSTRACT: Perchlorates are highly soluble anions and are used as ingredients in solid
rocket fuels, fireworks, missiles, batteries etc. The potential human risk of perchlorate expo-
sures includes effects on nervous system, inhibition of thyroid activity and mental retarda-
tion in infants. Various materials and techniques have been used to remove perchlorate from
drinking water. For light polluted water of perchlorate, adsorption seems to be one of the
most attractive, easiest, safest and cost effective physio-chemical treatment methods espe-
cially for drinking water. Rice husk, one of the major bi-products of rice milling industry can
be used as a low cost adsorbent for perchlorate removal.
Present study deals with adsorption of perchlorate using cationic modified rice husk by
optimizing various parameters like pH, adsorbent mass, adsorbate concentration, tempera-
ture and time of adsorption. The surface charge is the major governing factor for perchlo-
rate removal compared to surface area. To enhance the adsorption capacities, modifications
with cationic surfactants were made. Powdered rice husk was surface modified with Cetyl
Trimethyl Ammonium Bromide (CTAB). The adsorption of perchlorate was studied experi-
mentally after surface modifications and the different parameters including pH, adsorbent
mass, adsorbate concentration, temperature and time were optimized. The performance of
adsorption was found at different conditions and it has been observed that more than 97%
adsorption efficiency was achieved in the perchlorate removal.
1 INTRODUCTION
Perchlorate (ClO4−) is highly soluble anion that consists of a central chloride atom surrounded
by four oxygen atoms (John. D. Coates et al. 2004). Perchlorate salts have been manufactured
and used as ingredients in solid rocket fuels highway safety flares, airbag inflators, fireworks,
missile, fuels, batteries, matches (Mamie N.I. et al. 2014). It mostly exists as ammonium per-
chlorate, sodium perchlorate, potassium perchlorate, magnesium perchlorate and lithium
perchlorate (Yali Shi et al. 2007, Urbansky 1998). The perchlorate ion is similar in size to an
iodide ion and can therefore be taken up in place of iodide ions by the thyroid gland. Thus the
perchlorate ions disturb the production of thyroid hormones and may disrupt metabolism
in the human body and the effects can be significant in case of pregnant women and fetuses
(Urbansky 2002). The potential human risk of perchlorate exposures include effects on
nervous system, inhibition of thyroid activity and mental retardation in infants (Z. Li, et al.
2000) Various materials and techniques have been used to remove perchlorate from drinking
water. These technologies are classified physical removal by the sorption on materials, chemi-
cal reduction by metal, biodegradation by bacteria and electrochemical reduction on metal
electrodes and integrated techniques. Now a days the better method for the treatment of
549

wastewater is considered as adsorption due to its universal nature, inexpensiveness and ease
of operation (Guillame Darracq et al. 2014).
For light polluted water of perchlorate adsorption seems to be one of the most attractive, eas-
iest, safest, versatile and cost effective physio-chemical treatment methods, especially for drink-
ing water. Adsorption can also remove soluble and insoluble organic pollutants. The removal
capacity by this method may be up to 99.9%. Due to these facts, adsorption has been used for
the removal of a variety of organic pollutants from various contaminated water sources.
The biggest barrier in the application of this process by the industries is the high cost of
adsorbents presently available for commercial use. The cost of adsorption technology can be
reduced, if the adsorbent is in expensive. So there is a need to develop low cost and easily
available adsorbents for the removal of these pollutants from the aqueous environment. An
abundant source of potentially anionic compounds adsorbing biomass is cellulosic agricul-
tural waste. They are widely available, inexhaustible and inexpensive material that exhibit
specificity towards targeted ions (Umesh. K.G. et al, 2009).
The natural adsorbent can improve their adsorption capacity through various modifications.
The treatment with surfactant is one of the effective modifications. The cationic surfactant modi-
fied adsorbent alters the surface property of the adsorbent from hydrophilicity to hydrophobicity
or organophilicity. The modified adsorbent enhances their adsorption capacity for perchlorate
in water (Umpuch et al, 2013). Present work focused on modification of rice husk using cationic
surfactant CTAB for the better adsorption for perchlorate (Dipu Borah et al, 2009).
Recently several materials have been tested for perchlorate removal. A common adsorb-
ent used for the adsorption of perchlorate was granular activated carbon. To enhance the
adsorption capacities of granular activated carbon several modifications with cationic sur-
factants was made (Parette & Cannon, 2005). It was found that perchlorate adsorption on 10
types of commercial activated carbon followed Langmuir type of adsorption isotherm. The
surface charge was the most important factor governing the perchlorate removal rather than
specific surface area (Rovshan Mahmudov et al. 2010).
Rice husk is one of the major bye-product of the rice milling industry, is one of the most
commonly available lingo-cellulosic materials that can be converted to different types of fuels
and chemical feed stocks through a variety of thermochemical conversion processes. Chemi-
cal content of rice husk consists of 50% cellulose, 25–30% lignin and 15–20% silica (Humay-
atul Ummah et al. 2015).
2 MATERIAL AND METHODS
2.1 Adsorbent collection and preparation

Rice husk was used in the present work were collected from Kariavattom, Thiruvanan-
thapuram – Kerala – GPRS 8º 33’ 51” N, 76º 53’ 11” E. The collected Rice husk were washed
with distilled water several times to remove dirt particles and water soluble materials. The
washed materials were dried in an air oven at 40°C for 24 h. The dried Rice husk then pow-
dered and sieved to the desired particle size (40 µm). Finally the product was stored in a
vacuum desiccator until required.
2.2 Surface modification of rice husk using CTAB

2 g of adsorbent (rice husk) was treated with 20 ml of 1% CTAB (1 g in 100 ml) and shaken in a
temperature controlled shaker at 100 rpm for 24 h. The modified rice husk was then filtered and
washed with distilled water several times. It was then dried in hot air oven at 60°C overnight.
2.3 Adsorption experiment and analysis

2.3.1 Estimation of optimum pH for the experiment
The optimum pH for the adsorption, was found out by taking 2 g of modified rice husk in
a series of conical flask containing 10 mg/l of adsorbate. The pH was adjusted to 2–10 and
550

Figure 1. Overall procedure for measuring the residual perclorate.
kept for shaking. After 1 hr incubation the samples were filtered and analysed the residual
perchlorate.
2.3.2 Estimation of optimum amount of adsorbent

The optimum amount of rice husk powder for the adsorption was found out by weighing the
adsorbent ranging from 1 to 5 g in a series of conical flask containing 10 mg/l adsorbate. The
solution was under optimum pH condition and kept on the shaker for 1 hr. After incubation
the samples were filtered and residual perchlorate was analysed.
2.3.3 Estimation of optimum temperature

Optimum amount of modified rice husk was taken in a series of conical flask containing
10 mg/L adsorbate, and the solution was kept for shaking at temperatures ranging from
30°C to 70°C for 1 h. After incubation the samples were filtered and analysed the residual
perchlorate.
2.3.4 Estimation of optimum perchlorate concentration

In order to obtain the optimum perchlorate concentration, the experiment was carried out at
perchlorate concentration ranging from 5 to 25 ppm respectively; in a series of conical flask
containing optimum amount of modified adsorbent. The other parameters like temperature,
mass of adsorbent and pH condition were kept constant. After incubation, the samples were
taken for perchlorate analysis using the ion sensitive electrode.
2.3.5 Estimation of optimum level of perchlorate adsorption

In order to obtain the optimum level of perchlorate adsorption, the experiment was carried
out under optimum conditions of pH, adsorbent mass, adsorbate concentration and tem-
perature. The samples were kept for shaking and after 1 hr of incubation it was filtered and
residual perchlorate was analysed.
3.1 Adsorption at different pH

The effect of pH solutions on removal of perchlorate from aqueous solutions with range of
pH varying from 2 to10 is studied. Figure 2 shows the effect of pH on percentage removal of
551

Figure 2. Effect of pH on the removal of per- Figure 3. Effect of adsorbent mass on
chlorate by rice husk at 37 ± 2°C. removal of perchlorate by rice husk at 37 ± 2°C
and pH4.
perchlorate. It can be noted that the percentage removal of perchlorate increases and reaches
maximum at pH-4 and then decreases. This behavior can be due to the effect of pH solution
on the charge of functional group of rice husk, and thus become more effective to adsorption
in acidic pH. The modified rice husk containing a positive charge in this acidic pH is more
effective for the adsorption of maximum amount of perchlorate.
3.2 Adsorption at different adsorbent mass

The effect of adsorbent dosage on the removal percentage of perchlorate increases and reaches
maximum at 3 g of adsorbent and then decreased. Figure 3 shows the results obtained from
this investigation. This increase in the removal percentage with increase amounts of adsorb-
ent may be due to the availability of more effective sites of adsorption due to increased
surface area.
The decrease of adsorption with increasing the amounts of adsorbent can be attributed to
unsaturated absorption sites residual through adsorption procedure.
3.3 Adsorption at different adsorbate concentration

The effect of concentration of adsorbate on the removal percentage of perchlorate from
aqueous solutions with perchlorate concentrations ranging from 5 to 25 mg/L is studied.
Figure 4 shows the effect of adsorbate concentration on removal of perchlorate by rice husk
at 37 ± 2°C and keeping other parameters (pH-4, Adsorbent mass-3 g, adsorbate) constant.
The adsorption of perchlorate in rice husk is maximum at 5 ppm and the percentage reduc-
tion was also high at this level. The percentage adsorption rate increases with increase in
perchlorate concentration but the adsorption per unit gram of adsorbent reaches maximum
at 20 ppm and reaches equilibrium. The effect of the adsorbed concentration factor depends
on the relationship between the concentration of the adsorbed and the binding sites available
on surface of the adsorbent.
3.4 Adsorbtion at different temperatures

The effect of temperature on the removal percentage of perchlorate from aqueous solutions
with temperature varying between from 30°C to 70°C was studied. Figure 5 shows the effect
of temperature on removal of perchlorate by rice husk at (30–70°C) by keeping the parame-
ters (pH-4, Adsorbent mass-3°g, adsorbate concentration-20 ppm) constant. The adsorption
of perchlorate in rice husk is maximum at 60°C and the percentage reduction was also high
at this temperature. According to Le-Chatelier’s Principle, adsorption occurs more readily at
lower temperature and decreases with increase in temperature.
552

Figure 4. Effect of adsorbate concentration Figure 5. Effect of temperature on removal of
on removal of perchlorate by rice husk. perchlorate by rice husk at (30–70°C).
Figure 6. Effect of optimum parameters.
3.5 Adsorption at optimum parameters

The removal of perchlorate was investigated at time interval of 10 minutes. Figure 6: shows
the effect of optimum parameters (pH-4, temperature (60°C), adsorbent mass-3 g, concen-
tration 20 ppm) on removal of perchlorate by rice husk. Results showed that the removal
percentage increased and reached the equilibrium point at 60 minutes.
4 CONCLUSION
This study deduces that the cationic surfactant modified rice husk powder can be used as a
very low cost adsorbent for the removal of perchlorate contaminated water. From the results
obtained it can be concluded that rice husk is a good adsorbent due to its ability to remove
the perchlorate in water even at low concentrations. The experiment showed that at low pH
the adsorption level was maximum because CTAB modified rice husk containing a positive
charge which increases the adsorption rate of negatively charged.
REFERENCES
[1] John, D.C. & Laurie, A.A. 2004. Microbial perchlorate Reduction, Rocket fuelled metabolism.
Nature Reviews/Microbiology 2:569–580.
[2] Mamie, N.I., Kate, M.S. & Dennis, C.R. 2005. Reduction of Perchlorate and Nitrate by Microbial
Communities in Vadose Soil. Applied and Environmental Microbiology, 71(7):3928–3934.
553

[3] Yali, S., Ping, Z., Yawei, W., Jianbo, S., Yaqi, C., Shifen, M. & Guibin. 2007. Perchlorate in sewage
sludge, rice, bottled water and milk collected from different areas in China. Environment Interna-
tional 33:955–962.
[4] Sridhar, K. 2013. Organoclays of high-charge synthetic clays and alumina pillared natural clays:
Perchlorate uptake. Applied Clay Science 80:340–345.
[5] Urbansky, E.T. 2002. Perchlorate as an environmental contaminant. Environmental Science and
Pollution Research 9:187–192.
[6] Z. Li, F.X. Li, D. Byrd, G.M. Deyhle, D.E. Sesser, M.R. Skeels, S.H. Lamm. 2000 Neonatal thyrox-
ine level and perchlorate in drinking water.Occup.Environ.Med. 42:200–205.
[7] Robert, P. & Fred, S.C. 2005. The removal of perchlorate from ground water by activated carbon
tailored with cationic surfactants. Water Research 39:4020–4028.
[8] Rovshan, M. & Chin, P.H. 2010. Perchlorate removal by activated cabon asorption. Separation and
purification technology 70:329–337.
[9] Weifang, C., Fred, C. & Joserene, R.M. 2005. Ammonia-tailoring of Gac to enhance perchlorate
removal II Perchlorate Adsorption. Carbon 43(3):581–590.
[10] Yanhua, X., Shiyn, L., Fei, W. & Guangli, L. 2010. Removal of perchlorate from aqueous solution
using protonated cross-linked chitosan. The Chemical Engineering Journal 156(1):56–63.
[11] Muquing, Q., Chu, X.H., Xiaojie, L. 2016. The influence of the co-existing anions on the adsorp-
tion of perchlorate from water by the modified orange peels. Mature Environment and Pollution
Technology 15 (4):1359–1362.
[12] Joo, Y.K., Sridhar, K., Robert, P., Fred, C., Hiroaski, K. 2011. Perchlorate uptake by synthetic
layered double hydroxides and organo-clay materials. Applied Clay Science 51:158–164.
[13] Umesh, K.G., Kaur, D.S. & K. Garg. 2009. Removal of hexavalent chromium from aqueous solu-
tion by adsorption on treated sugarcane bagasse using response surface methodological approach.
Desalination 249:475–479.
[14] C. Umpuch & B. Jutarat. 2013. Adsorption of Organic Dyes from Aqueous solution by Surfactant
Modified Corn Straw. International Journal of Chemical Engineering and Applications 4(3).
[15] Dipu, B., Shigeo, S., Shigeon, K. & Toshinri, K. 2009. Sorption of As(V) from aqueous solution
using acid modified Carbon black. Journal of Hazardous Materials 162:1269–1277.
[16] Humayatul, U., Dadang, A.S., Mary, S. & Abdul, W. 2015. Analysis of Chemical Composition of
ice Husk Used as Absorber plates sea water into clean water, ARPN Journal of Engineering and
Applied Sciences 10(14):6046–6050.
[17] Farai, M., Olga, K. & Pardon, K.K. 2014. Removal of Cr(VI)from aqueous solutions using Powder
of Potato peelings as a low cost sorbent, Hindawi Publishing Corporation.
554

Biosorption of methyl orange from aqueous solution using

cucurbita pepo leaves powder
M. Tukaram Bai, K. Latha, P. Venkat Rao & Y.V. Anudeep

Department of Chemical Engineering, Andhra University, Visakhapatnam, India
ABSTRACT: This paper investigates the biosorption of methyl orange from aqueous
solution on cucurbita pepo leaves powder. The batch studies were carried out for the con-
tact time, (5–60 min), biosorbent dosage (0.05–0.5 g), pH (2–9), initial concentration of dye
(10–50 mg/L) and temperature, (283–323°K). The isotherms for the present study are Freun-
dlich, Langmuir and Temkin. Of these isotherms, the Freundlich isotherm was best fitted.
The kinetics were studied with pseudo first and second order. Biosorption kinetics of cucur-
bita pepo leaves were well correlated with pseudo second order.
Keywords: biosorption, methyl orange, cucurbita pepo leaves, effluent treatment, isotherms,
kinetics
1 INTRODUCTION
The development of science and technology provides many benefits to human life, but for
the effects of a negative impact on the surrounding environment, such as industrial waste
problems. Many industries use dyes in order to color their products, and then dispose of a
lot of colored wastewater as effluent. The major sources of the dyes are industries like car-
pet, leather, printing, textile. The wastewater from these industries causes hazardous health
effects to aquatic and human life.
Several processes have been applied for the treatment of dyes from wastewater, such as
chemical, biological and physical processes. Even though chemical and biological treatments
are effective for removing dyes, they still require special equipment and are considered to
be quite energy intensive in terms of the addition and large amounts of by products often
generated. In recent years, a physical method through adsorption process based on acti-
vated carbon material has been considered to be a superior technique as compared to others.
However, the commercial activated carbon is quite expensive and has limited its application.
Due to economic reasons, the discovery toward alternative adsorbents to replace the costly
activated carbon is highly recommended. Many investigators have studied the feasibility of
using inexpensive alternative materials like chitosan beads (Negrulescu et al., 2014), calcined
Lapindo (Jalil et al., 2010), chitosan intercalated montmorillonite (Umpuch & Sakaew, 2013),
activated carbon coated monolith in a batch system (Darmadi & Thaib, 2010), sawdust and
sawdust-fly ash (Lucaci & Duta, 2011), cork as a natural and low-cost adsorbent (Krika &
Benlahbib, 2015), thermally treated eggshell (Belay & Hayelom, 2014), modified activated
carbon from rice husk (Qiu et al., 2015), tree bark powder (Egwuonwu, 2013), and banana
trunk fiber (Prasanna et al., 2014), as carbonaceous precursors for the preparation of acti-
vated carbons and for the removal of dyes from water and wastewater. The present investiga-
tion is an attempt to explore the possibility of using cucurbita pepo leaves powder to remove
methyl orange in aqueous solution, since the raw material is harmless, cheaper, and plentiful.
Methyl orange is an anionic azo dye with a molecular formula C14H14N3NaO3S. The wide
usage of methyl orange was as a pH indicator. It is used in titrations because it changes
color at the pH of a mid-strength acid. Its anion form is yellow, and its acidic form is red.
555

The entire color change occurs in acidic conditions. It is partially soluble in hot water,
very slightly soluble in cold water, and insoluble in diethyl ether and alcohol. It is soluble
in pyrimidine. At room temperature it appears as a solid, odorless, orange/yellow powder
that yields an orange solution when dissolved in water. The IUPAC name is Sodium 4-[(4-
dimethylamino) phenyldiazenyl] benzenesulfonate.
Its oral consumption is toxic, hazardous and carcinogenic due to the presence of nitro-
gen. When discharged into the running streams it will affect aquatic life, causing detrimental
effects in the liver, gill, kidney, intestine, and pituitary gonadotropic cells. In humans, it may
cause irritation to the respiratory tract if inhaled and causes irritation to the gastrointestinal
tract upon ingestion. This dye may enter into the food chain and could possibly cause carci-
nogenic, mutagenic, and tetragenic effects on humans.
2.1 Preparation of biosorbent

The cucurbita pepo leaves used in the present study were collected from the plants near
Gopalapatnam, Visakhapatnam. The collected leaves were washed twice with distilled water
to remove dirt, completely dried in sunlight for 20 days, and then dried in an oven to remove
the excess moisture content. The dried leaves were then cut into small pieces and powdered.
The powder was screened to 150 µm size and directly used as a biosorbent.
2.2 Procedure
The methyl orange was obtained from Merck laboratories limited, Mumbai, India. A stock
solution of 1,000 mg/L was prepared by dissolving 1g of methyl orange in 1,000 mL of
distilled water, which was later diluted to required concentrations. All the solutions were
prepared using distilled water. Solution pH for pH studies was adjusted by adding HCl and
NaOH, as required. Concentrations of the dye solutions were determined from the absorb-
ance spectrum of the solution at the characteristic wavelength of dye using a double beam
UV-Visible spectrophotometer. Final concentrations were determined from the calibration
curve. The absorption wavelength of methyl orange (λmax) = 464 nm. Variables studied and
their range: Contact time: 5–60 min., Aqueous dye solution pH: 2–9, Initial concentration of
the dye: 10–50 mg/L, Biosorbent dosage: 0.05–0.2 g, Temperature: 283–323°K.
The effect of contact time was determined by shaking 0.1 g of adsorbent in 100 ml of syn-
thetic solutions methyl orange of initial dye concentration 10 mgL−1. Shaking was provided
for different time intervals like 5, 10, 15, 20, up to 60 min at a constant agitation speed of
230 rpm. The effect of pH of the dye solution was determined by agitating 0.1 g of biosorb-
ent and 100 ml of synthetic dye solutions of initial dye concentration 10 mg/L at different pH
values of the solution, ranging from 2 to 9, by adding 0.1N HCl or 0.1 N NaOH. To study
the effect of concentration, 100 ml of aqueous solution, each of methyl orange of different
dye concentrations of 10 mg/L, 20 mg/L, 30 mg/L, 40 mg/L and 50 mg/L were taken in 250 ml
conical flasks. 0.1 g/L of cucurbita pepo leaves powder was added to each of the flasks. The
total dye concentration in solution was analyzed with double beam a UV Spectrometer at a
wavelength of 464 nm for methyl orange dye solution.
Figure 1. Molecular structure of methyl orange.
556

Experimental data was generated in a batch mode of operation to study the effect of various
parameters for the removal of methyl orange from the aqueous solution (prepared in the
laboratory) using cucurbita pepo leaves powder as the biosorbent. Various experimental runs
were conducted in the present study. The parameters studied include: Contact time, t (min),
pH of the solution, initial concentration of the solution, C0 (mg/L), biosorbent dosage, w (g)
and temperature, T (°K).
3.1 Effect of contact time, t

The effect of contact time on biosorption of methyl orange onto cucurbita pepo was studied
at 10 mg/L. From Figure 2, it can be seen that the rate of biosorption was very rapid in the
initial period of contact time. Thereafter, it gradually increased with contact time until bio-
sorption was reached at the equilibrium point. This trend of biosorption kinetics was due to
the biosorption of dye on the exterior surface of biosorbent at the initial period of contact
time. When the biosorption on the exterior surface reached saturation point, the dye diffused
into the pores of the biosorbent and was adsorbed by the interior surface of the biosorbent.
The equilibrium time for the cucurbita pepo-methyl orange system was 40 min. No further
biosorption occurred, even after an increase in the contact time from 40 min to 60 min. The %
removal and dye uptake were calculated, as given in Equations 1 and 2 respectively:
% Removal =
( Co − Ct ) × 100 (1)
Co
Dye uptake ( Q ) =
( Co − Ct ) × v (2)
w × 1000
where
C0 = Initial concentration of the dye, mg/L
Ct = Final concentration of the dye after time, t
V = Volume of aqueous dye solution, ml
W = Weight of biosorbent, g
3.2 Effect of initial concentration of aqueous dye solution (C0)

Experiments were undertaken to study the effect of the initial dye concentration on the
removal of methyl orange from the solution. The variations of % dye removal and dye uptake
Figure 3. Effect of initial concentration of

methyl orange dye on % removal and dye uptake.
Figure 2. Effect of contact time on % removal of

methyl orange dye.
557

with initial dye concentration are presented in Figure 3. The obtained curves show that the
dye uptake increased with an increase in initial concentration of dye, while the percentage
removal of dye decreased with an increase in initial dye concentration. The increase in dye
uptake is a result of the increase in the driving force; that is, the concentration gradient, with
an increase in the initial dye concentration (from 10 to 50 mg/L). However, the percentage
removal of dye on cucurbita pepo leaves powder was decreased from 91.50 to 86.54% for
methyl orange. Though an increase in dye uptake was observed, the decrease in percentage
removal may be attributed to lack of sufficient surface area to accommodate much more
dye available in the solution. The percentage removal at higher concentration levels shows a
decreasing trend whereas the equilibrium uptake of dye displays an opposite trend.
3.3 Effect of pH
The pH of an aqueous dye solution is an important monitoring parameter in biosorption, as
it affects the surface charge of the biosorbent material and the degree of ionization of the dye
molecule. It is also directly related to the competition ability of hydrogen ions with biosorbate
molecules to active sites on the biosorbent surface. In the present study methyl orange dye bio-
sorption data was obtained in the pH range of 2 to 9 of the aqueous solution (C0 = 10 mg/L)
using 0.1 g of 150 µm size biosorbent. The effect of pH of aqueous solution on % biosorption of
methyl orange dye is shown in Figure 4. The % biosorption of methyl orange dye was increased
from 87.90 to 93.68% as pH increased from 2 to 6, and beyond the pH value of 6 it decreased.
As the pH of the system decreased, the number of negatively charges surface sites decreased
and the number of positively charged surface sites increased and this favors the biosorption
of dye anions due to electrostatic attraction. When the acidity increased due to concentration
of H+ that will decrease the negative charge for methyl orange, then the adsorption increased.
However, when the basicity of solution was increased, the amount of biosorption decreased due
to the increase of concentration of OH. Hence the optimum pH for methyl orange is taken as 6.
3.4 Effect of biosorbent dosage (w)

The percentage biosorption of methyl orange dye is drawn against a biosorbent dosage for
150 µm biosorbent size is shown in Fig. 5. The % biosorption of methyl orange dye increased
from 88.10 to 95.80 and dye uptake decreased from 17.62 to 4.765 mg/g, with an increase in
biosorbent dosage from 0.05 to 0.2 g. The number of active sites of biosorbent increase with
an increase in biosorption dosage. This is due to the binding of almost all the dye molecules
to adsorbent surfaces and the establishment of equilibrium between the dye molecules on the
adsorbent surfaces and in the solution.
Figure 5. Effect of biosorbent dosage on %

removal of methyl orange dye.
Figure 4. Effect of pH on % removal of methyl
orange dye.
558

Figure 6. Effect of temperature on % removal of methyl orange dye.
3.5 Effect of temperature (T)

The effect of change in the temperature on the methyl orange dye uptake is shown in Figure 6.
The biosorption of methyl orange by cucurbita pepo leaves powder at different temperatures
showed an increase in the biosorption capacity when the temperature was increased. The
temperature has two main effects on the biosorption process. An increase in temperature is
known to increase the diffusion rate of the biosorbate across the external boundary layer and
within the pores. Furthermore, changing the temperature will modify the equilibrium capac-
ity of the biosorbent for a particular biosorbate. The effect of temperature was investigated
from batch experiments carried out at five constant temperatures: 283, 293, 303, 313 and
323°K. With an increase in temperature, the % removal was increased from 89.73 to 94.18%
for methyl orange for the initial concentration of 10 mg/L, as shown in Figure 6. This indi-
cates that the biosorption reaction is endothermic in nature because of the chemical interac-
tion between biosorbate and the biosorbent, and due to the increased rate of intra-particle
diffusion of dyes into the pores of the biosorbent at higher temperatures. However, the dye
uptake capacity was increased with an increase in temperature.
4 BIOSORPTION KINETICS
The kinetics of the biosorption data was analyzed by two models, namely pseudo first order
and pseudo second order. These models correlate solute uptake, which is important in the
prediction of reactor volume.
4.1 Pseudo first order kinetics

The order of biosorbate—biosorbent interactions have been described using kinetic models.
Traditionally, the Lagergren first order model finds wide application. The Lagergren first
order rate Equation is:
(dQ/dt) = K1 (Qeq – Q) (3)
where Qeq and Q are the amounts of dyes adsorbed at equilibrium time and any time t, and
K1 is the rate constant of the pseudo first order biosorption.
The above Equation can be presented as:
∫ (dQ/(Qeq – Q)) = ∫ K1 dt (4)
Applying the initial condition qt = 0 at t = 0, we get:
559

Figure 7. First order kinetics for % biosorp-
tion of methyl orange dye. Figure 8. Second order kinetics for % biosorp-
tion of methyl orange dye.
log (Qeq – Q) = log Qeq – (K1/2.303) t (5)
The plot of time t versus log (Qeq–Q) gives a straight line for first order kinetics, facilitating
the computation of biosorption first order rate constant (K1).
In the present study, the kinetics were investigated with 100 ml of aqueous solution
(C0 = 10 mg/L) for the contact time of 5 to 60 min. The Lagergren first order plot is drawn in
Figure 7. The first order model Equation obtained for the present study is:
log (Qeq−Q) = −0.012t − 0.259 (6)
4.2 Pseudo second order kinetics

The pseudo second order model is represented by the following Equation:
(t/Q) = (1/K2Qeq2) + (1/Qeq ) t (7)
If the pseudo second order kinetics is applicable, the plot of time t versus (t/Q) gives a
linear relationship that allows computation of K2.
In the present study, the kinetics are investigated with 100 ml of aqueous solution (C0 = 10
mg/L) in the agitation time intervals of 5 min to 60 min. The pseudo second order plot of time
‘t’ versus (t/Q) is drawn in Figure 8. The second order kinetics obtained for the present study is
given as:
t/Q = 0.1075 t + 0.0447 (8)
5 ADSORPTION ISOTHERMS
In the present study, the isotherms studies are Langmuir, Temkin and Freundlich. The linear
forms of these isotherms are obtained at room temperature and are shown in the Figures below.
5.1 Langmuir isotherm

The Langmuir isotherm has been successfully applied to many pollutant biosorption processes
and has been the most widely used isotherm for the biosorption of a solute from a liquid solution.
560

A basic assumption of the Langmuir theory is that biosorption takes place at specific
homogeneous sites within the biosorbent. It is then assumed that once a solute particle occu-
pies a site, no further biosorption can take place at that site. The rate of biosorption to the
surface should be proportional to a driving force. The driving force is the concentration in the
solution, and the area is the amount of bare surface.
The Langmuir relationship is hyperbolic, and the Equation is shown as:
Q max bC eq
Qeq = (9)
1 + bC eq
The above Equation can be rearranged into the following linear form as:
C eq 1 1
= + C eq (10)
q eq bQ max Q max
where
Ceq is the equilibrium concentration (mg/L)
Qeq is the amount of dye ion adsorbed (mg/g)
Qmax is Qeq for a complete monolayer (mg/g)
b is sorption equilibrium constant (L/mg)
Figure 9 is the plot of [Ceq] versus [Ceq/Qeq], which is a straight line with slope 1/Qmax and
an intercept of 1/bQmax.
The Correlation coefficient R2 = 0.9671 and the Langmuir Equation obtained for the
present study is:
(Ceq/Qeq) = 0.0107Ceq + 0.0874 (11)
5.2 Freundlich isotherm

In 1906, Freundlich studied the sorption of a material onto animal charcoal. He found that if
the concentration of solute in the solution at equilibrium, Ceq, was raised to the power of m,
the amount of solute adsorbed being Qeq, then Ceqm/Qeq is a constant at a given temperature.
This fairly satisfactory empirical isotherm can be used for non-ideal sorption and is expressed
by the following Equation:
Figure 9. Langmuir isotherm for % biosorption of Methyl Orange dye.
561

Qeq = KfCeq1/m (12)
The Equation is conveniently used in the linear form by taking the logarithm of both sides as:
log (Qeq) = (1/m)* log (Ceq) + log (Kf) (13)
The Freundlich isotherm is derived assuming a heterogeneity surface. Kf and m are indica-
tors of biosorption capacity and biosorption intensity respectively. The value of m should lie
between 1 and 10 for favorable biosorption.
Figure 10 is a plot of log [Ceq] versus log [Qeq], which is a straight line with a slope of 1/m
and an intercept of log (Kf).
From the value of biosorption intensity, it can be concluded that the Freundlich isotherm
indicates for favorable biosorption. The Freundlich Equation obtained for the present study
is shown by:
log Qeq = 0.7454log Ceq + 1.0227 (14)
Figure 10. Freundlich isotherm for % biosorption of methyl orange dye.
Figure 11. Temkin isotherm for % biosorption of methyl orange dye.
562

5.3 Temkin isotherm
The derivation of the Temkin isotherm assumes that the fall in the heat of sorption is lin-
ear rather than logarithmic, as implied in the Freundlich Equation (Aharoni & Ungarish,
1977). The Temkin isotherm has generally been applied in the following form:
R *T
Qeq = ln[AT * Ceq ] (15)
bT
where
R = Universal gas constant (8.314 J/mol.K)
T = Temperature of dye solution, K
AT, bT = Temkin isotherm constants
This can be written as:
R *T R *T
Qeq = ln  Ceq  + ln [ AT ] (16)
bT bT
Figure 11 shows a plot of ln[Ceq] versus Qeq, which is a straight line with slope of RT/bT
and intercept of RT/bT ln[AT]. The Temkin Equation obtained for the present study is:
Qeq = 0.0598ln Ceq − 0.5235 (17)
The isotherm constants obtained for various isotherm models are shown in Table 1.
The correlation coefficients obtained from the Langmuir, Freundlich, and Temkin models
were 0.9671, 0.9981, 0.959 respectively for methyl orange, and the Freundlich Equation was
observed to be more suitable, followed by Langmuir and Temkin for the experimental data
of methyl orange dye.
The maximum metal uptakes for the biosorption of the methyl orange by using various
other biosorbents are tabulated in Table 2.
Table 1. Isotherm constants for various isotherm models.
ISOTHERM CONSTANT R2
Langmuir Qmax, mg/g = 93.45 0.9671

b, L/mg = 0.1224
Freundlich Kf = 10.536 0.9981
m = 1.342
Temkin AT = 1.57 × 10-4 0.959
bT = 42,126
Table 2. Methyl orange uptake capacities for different biosorbents.
AUTHOR BIOSORBENT Qmax, mg/g
Jalil et al. (2010) Calcined Lapindo volcanic mud 333.3

Darmadi and Tahib (2010) Carbon coated monolith 36.72
Krika and Benlahbib (2015) Cork powder 16.66
Faith Deniz (2013) Prunus amygdalus (almond shell) 41.34
Danish et al. (2013) Acacia mangium wood 7.54
Gong et al. (2013) Finger citron residue 934.58
Su et al. (2014) Wheat straw 50.4
Chaidir et al. (2015) Durio zibethinus (murr seed) 6.352
Present study Cucurbita Pepo 93.45
563

6 CONCLUSION
The following conclusions can be drawn from the above discussion. Methyl orange is removed
efficiently by the cucurbita pepo leaves. The optimum contact time for the process is 40min
at room temperature. The optimum pH is 6 and the optimum dosage is 0.1g. The biosorp-
tion favors the increase in the temperature. The adsorption kinetics are better described with
pseudo second order kinetics. The isotherm studies are best fitted for the Freundlich iso-
therm, followed by the Langmuir and Temkin isotherms.
REFERENCES
Belay, K. & Hayelom, A. (2014). Removal of methyl orange from aqueous solutions using thermally
treated egg shell (locally available and low cost biosorbent). International Journal of Innovation and
Scientific Research, 8(1), 43–49. ISSN 2351-8014.
Chaidir, Z., Sagita, D.T., Zein, R & Munaf, E. (2015). Bioremoval of methyl orange dye using durian
fruit (durio zibethinus) murr seeds as biosorbent. Journal of Chemical and Pharmaceutical Research,
7(1), 589–599.
Danish, M., Hashim, R., Ibrahim, M.N.M. & Sulaiman, O. (2013). Characterization of physically acti-
vated acacia mangium wood-based carbon for the removal of methyl orange dye. BioResources, 8(3),
4323–4339.
Darmadi, D. & Thaib, A. (2010). Adsorption of anion dye from aqueous solution by activated carbon
coated monolith in a batch system. Jurnal Rekayasa Kimia dan Lingkungan, 7(4), 170–175. ISSN
1412–5064.
Deniz, F. (2013). Adsorption properties of low-cost biomaterial derived from Prunus amygdalus L. for
dye removal from water. The Scientific World Journal, 961671. Egwuonwu, P.D. (2013). Adsorption
of methyl red and methyl orange using different tree bark powder. Academic Research International,
4(1), 330.
Gong, R., Ye, J., Dai, W., Yan, X., Hu, J., Hu, X., & Huang, H. (2013). Adsorptive removal of methyl
orange and methylene blue from aqueous solution with finger-citron-residue-based activated carbon.
Industrial & Engineering Chemistry Research, 52(39), 14297–14303.
Jalil, A.A., Triwahyono, S., Adam, S.H., Rahim, M.D., Aziz, M.A.A., Hairom, N.H.H., &
Mohamadiah, M.K.A. (2010). Adsorption of methyl orange from aqueous solution onto calcined
Lapindo volcanic mud. Journal of Hazardous Materials, 181(1–3), 755–762.
Krika, F. & Benlahbib, O.E.F. (2015). Removal of methyl orange from aqueous solution via adsorp-
tion on cork as a natural and low-cost adsorbent: Equilibrium, kinetic and thermodynamic study of
removal process. Desalination and Water Treatment, 53(13), 3711–3723.
Lucaci, D. & Duta, A. (2011). Removal of methyl orange and methylene blue dyes from wastewater
using sawdust and sawdust-fly ash as sorbents. Environmental Engineering and Management Journal,
10(9), 1255–1262.
Negrulescu, A., Patrulea, V., Mincea, M., Moraru, C. & Ostafe, V. (2014). The adsorption of tartrazine,
congo red and methyl orange on chitosan beads. Digest Journal of Nanomaterials and Biostructures,
9(1), 45–52.
Prasanna, N, Manivasagan, V., Pandidurai, S., Pradeep, D. & Leebatharushon, S.S. (2014). Studies on
the removal of methyl orange from aqueous solution using modified banana trunk fibre. Interna-
tional Journal of Advanced Research, 2(4), 341–349.
Qiu, M.Q., Xiong, S.Y., Wang, G.S., Xu, J.B., Luo, P.C., Ren, S.C. & Wang, S.B. (2015). Kinetic for
adsorption of dye methyl orange by the modified activated carbon from rice husk. Advance Journal
of Food Science and Technology, 9(2), 140–145. ISSN: 2042-4868, e-ISSN: 2042-4876.
Su, Y., Jiao, Y., Dou, C. & Han, R. (2014). Biosorption of methyl orange from aqueous solutions
using cationic surfactant-modified wheat straw in batch mode. Desalination and Water Treatment,
52(31–33), 6145–6155.
Umpuch, C. & Sakaew, S. (2013). Removal of methyl orange from aqueous solutions by adsorption
using chitosan intercalated montmorillonite. Songklanakarin Journal of Science & Technology, 35(4),
451–459.
564

Catalytic hydrodechlorination of 1,4-dichlorobenzene

from wastewater
C. Megha
Government Engineering College, Kozhikode, India
K. Sachithra
Sanjay P. Kamble
National Chemical Laboratory, Pune, India
ABSTRACT: Dechlorination is one of the promising methods to convert more toxic chlo-
rinated aromatics into less toxic environmentally friendly value-added products. In this work
catalytic hydrodechlorination is achieved with the aid of low-cost hydrogenation catalysts
such as Raney nickel, bimetallic catalyst and palladium on activated carbon catalyst. Among
them, Raney nickel shows the best result as it takes only a few hours to completely dechlo-
rinate 1,4-dichlorobenzene to benzene. The detailed study on dechlorination using Raney
nickel was made in this work. It is an economically feasible method for treating chlorinated
pollutants in wastewater. Experiments were done with varying parameters like temperature,
concentration, and pH. Dechlorination shows the best result at high temperature and lower
pH. The product was confirmed using HPLC analysis and UV spectroscopy. Recycling of
Raney nickel catalyst was also performed in this study. It was not possible to do more than
two recycles due to poisoning of the catalyst.
Keywords: Dechlorination, hydrodechlorination, 1,4-dichlorobenzene, wastewater
1 INTRODUCTION
Chlorinated aromatic compounds have at least one chlorine atom covalently attached to an
aromatic ring. Due to the presence of a halogen group in the aromatic ring, it is highly resist-
ant for biodegradability, and so its ubiquitous presence can be seen in every ecosystem. Chlo-
rinated organic, such as Trichloroethylene (TCE), Carbon Tetrachloride (CT), chlorophenols,
and Polychlorinated Biphenyls (PCBs) are among the most common contaminants. Most of
these chloroorganics were widely used in industry during the past half-century as solvents,
pesticides, and electric fluids.
Many chlorinated organic chemicals (COCs) have been detected in many surface waters
and groundwater, in sewage, and in some biological tissues (Pearson, 1982). The observed lev-
els are, in general, too low to cause immediate acute toxicity to mammals, birds, and aquatic
organisms (Cheng et al., 2007). Treatment processes can and do reduce the concentrations of
COCs in water However, the degree of efficacy is often a function of chemical structure, cost,
and energy. All treatment processes have some degree of side effects, such as a generation of
residuals or by-products. Among the different methods of treatment, catalytic hydrodechlo-
rination is emerging as an effective way to reduce toxicity of COCs. This method reduces the
toxicity and increases the biodegradability at low cost. It is a method of recycling compounds
from which they have originally formed with low emissions.
565

COCs, including 1,4-dichlorobenzene, have been reported as being primary pollutants by
Environmental Protection Agency. Several research studies revealed reducing the toxicity of
compound by breaking carbon chlorine bond is not effectively done. The greatest challenge
in implementing this strategy is the adoption of low-cost technologies using different metal
and bimetallic catalysts.
Various metal, bimetallic, hydrogenation catalysts were used. New bimetals were prepared
(Sopoušek et al., 2014) and dechlorination was done to convert 1,4-dichlorobenzene into ben-
zene. Screening of 1,4-dichlorobenzene was done using various metal and bimetallic catalysts.
Hydrogenation catalysts like Raney nickel, and palladium on activated carbon, were used to
dechlorinate 1,4-dichlorobenzene at room temperature and pressure (Xia et al., 2009).
2 EXPERIMENT
Experiments were conducted for studying the effect of initial concentration of 1,4-
dichlorobenzene, catalyst concentration, pH, and synergy of salts for Raney nickel. Dechlo-
rination experiments were conducted with various metal and bimetallic catalysts. A bimetallic
catalyst was prepared using its corresponding metal precursor and sodium borohydride was
used as a reducing agent. All experiments were carried out at room temperature (30 ± 2°C).
Stock solution of 1,000 ppm 1,4-dichlorobenzene was prepared in a 100 mL standard flask
with acetonitrile. The desired concentration of 1,4-dichlorobenzene for experiments was pre-
pared by micro-pipetting from this stock solution into deionized water.
The solutions of 1,4-dichlorobenzen were prepared in the concentrations of 10, 20, 50,
and 60 mg/L. Batch experiments were conducted with 150 mL of solution taken in a 250 mL
conical flask with a tight lid. These bottles were kept in a rotary shaker at a constant shaking
speed of 150 rpm for 24 hrs. After 24 hrs, 10 mL of sample was taken out and syringe fil-
tered. The resulting 1,4-dichlorobenzene concentration was determined using HPLC. From
this final 1,4-dichlorobenzene concentration, the percentage dechlorination was calculated,
% Dechlorination = (C−C0)/C0 *100 (1)
where C (mgL−1) is the amount of 1,4-dichlorobenzene per liter at time t., and C0 is the
initial concentration (mg L−1). The dechlorination efficiency or conversion percentage of 1,4-
dichlorobenzene was calculated using the expression:
2.1 Kinetic experiment study

In order to estimate the rate constant and order of reaction for the dechlorination of
1,4-dichlorobenzene, time-dependent studies were conducted in a 250 mL round-bottom
flask. Stirring was done continuously. The particular concentration of 1,4-dichlorobenzene
was transferred into the flask and optimized catalyst loading (3 g/L) of the catalyst was added
to it. Samples were withdrawn from the vessel at frequent time intervals and analyzed for the
concentration of 1,4-dichlorobenzene using HPLC. The kinetics were checked, mainly for
first-order reaction as most of the catalytic dechlorination reactions followed it.
ln C/C0 = −K*t (2)
where
C: concentration of 1,4-dichlorobenzene in ppm
C0: initial. Concentration of 1,4-dichlorobenzene in ppm
K: reaction rate constant
t:time
After a time interval, solutions were filtered and analyzed for 1,4-dichlorobenzene content
using HPLC.
566

3.1 Calibration of 1,4-dichlorobenzene

1,4-dichlorobenzene is calibrated in HPLC before any kind of analysis is done. 5, 10, 20, and
50 ppm 1,4-dichlorobenzene solution is prepared and kept for analysis.
Linear fit:
ax + b
a = 5.98309 × e-0.005
b = 0
Goodness of fit is 0.994
3.2 Catalysts screening

The dechlorination experiments of 1,4-dichlorobenzene were conducted with various metals,
bimetal, and metal nanoparticle with optimized catalyst loading 3g/L. Sodium borohydride
was used as a reducing agent.
The results show that hydrogenation catalysts like Raney nickel and palladium on acti-
vated carbon give the best results at room temperature and pressure. Fe/Cu bimetal also shows
Figure 1. Calibration curve of 1,4-dichlorobenzene.
Figure 2. Percentage dechlorination of various catalysts.
567

better dechlorination results. Economically Raney nickel is the feasible catalyst. Low-cost
bimetals were prepared, and dechlorination of 1,4-dichlorobenzene. using catalysts like Fe/Cu,
Al/Ni, Cu/Ni, and K-10 were reported for the first time. However, it gives comparatively poorer
results than the hydrogenation catalyst due to leaching and blockage of the bimetal surface.
3.3 Effect of catalyst loading

Dechlorination of 1,4-dichlorobenzene was done using a Raney nickel catalyst of various
loadings of 0.1 g, 0.3 g, 0.5 g, 0.7 g, and 0.9 g, keeping other parameters like initial concen-
tration, pH, and temperature constant. The catalyst loading is the most important parameter
in optimization of reaction conditions, which shows that as catalyst loading increases, the
dechlorination also increases. The activity of the catalyst is increased in an increased amount
as hydrogenation increases with loading. The screening was done with these loadings in a
rotary shaker for six hours and samples were analyzed in HPLC. At 3.3 gm/l of catalyst load-
ing, 99.9% dechlorination takes place.
3.4 Concentration screening

The concentration of reactant is an important parameter in dechlorination. The concentra-
tion is varied, such as 10 ppm, 20 ppm, 50 ppm, and 60 ppm, keeping the catalyst loading
at 3 g/L at room temperature. As concentration is increased the dechlorination is decreased.
Lowering the concentration too much also decreases the dechlorination. It has been observed
that between 20 ppm and 50 ppm dechlorination reach 98%. When the concentration is high
dechlorination is reduced; it may be due to agglomeration of more compounds in catalyst.
Therefore, the reaction would not takes place in a good manner. The only reactant in this
reaction is 1,4-dichlorobenzene. Thus, lowering the concentration will reduce the rate of
reaction; in effect, it decreases the dechlorination.
3.5 Effect of salts on dechlorination

Salts like ammonium sulfate, sodium carbonate, magnesium chloride, and sodium nitrate are
very common in ground water and other sources of water. It was observed that, except for
sodium nitrate salt, other salts have not much synergic effect on dechlorination. As observed,
dechlorination kinetics is dominant in the first hour. So, the sample after one hour was taken
to check the effect of various salts on dechlorination. The samples without salts showed
better dechlorination of the same initial concentration than with salts. The salts may affect
Figure 3. Catalyst loading.
568

the hydrogenation in such a way that the ions of the salts may block the catalyst sites, and
dechlorination efficiency is reduced comparatively. Sodium nitrate shows better dechlorina-
tion compared to without salt. This may be due to the presence of nitrate group in it, which
enhances the reduction reaction.
3.6 Effect of pH
pH is an important parameter in dechlorination because the ground water for treatment
will have different pH, due to different treatment and the presence of several salts and other
chemicals. Most of dechlorination studies show that acidic pH favors the dechlorination.
The experiments had been done with pH 2, 7, and 11 to know whether acidic, neutral or
basic pH is good for dechlorination. pH was changed using 0.1 N HCl and 0.1 N NaOH.
Figure 6 shows that acidic medium favors dechlorination in the first ten minutes as the sup-
ply of hydrogen ions from hydrochloric acid helps the fast formation of benzene, more than
in a basic condition where it is difficult for the replacement of hydrogen ions from water to
the aromatic ring. The result shows that acidic and neutral pH does not show much differ-
ence in dechlorination after four hours of reaction time. However, basic pH does not show
comparatively good results. Therefore, optimization of dechlorination can be done at neutral
pH, which is economically and chemically effective.
3.7 Effect of temperature

Studies based on the effect of temperature on dechlorination reaction were conducted at
various temperatures, such as at 40°C and 50°C. The studies show that as reaction tempera-
ture increases the dechlorination also increases. At 50°C, 86.77% dechlorination is achieved
while at 40°C only 37.54% dechlorination is achieved. The reaction was conducted only to a
temperature of 50°C because mild conditions are referred for COCs.
Figure 4. Concentration screening.
Figure 5. Effect of salts.
569

3.8 Effect of reducing agent
Reducing agent in a reaction mixture is a provider of hydrogen. It enhances the hydrodechlo-
rination reaction to a great extent. The dechlorination reaction is conducted without reduc-
ing agent and with reducing agent. Sodium borohydride is used as a reducing agent and
Raney nickel is used as a catalyst at a loading of 3 gm/L. Reaction is conducted for two hours
and samples are analyzed for the different time intervals. While using Raney nickel as a cata-
lyst, usage of reducing agent can be avoided. From Figure 8 we can interpret that dechlorina-
tion is almost the same for with and without reducing agent.
3.9 Reducing agents

Reducing agent is a proton donor or hydrogen donor, so hydrodechlorination reaction will
occur in a good manner. Different reducing agents like sodium borohydride, oxalic acid,
2-propanol, sodium nitrite, and activated charcoal were used for the comparative study.
Reaction was kept for 24 hours. Sodium nitrite showed 99.5% dechlorination and sodium
borohydride showed 96.45%. Considering the economic side, sodium borohydride was best.
Figure 6. Effect of pH.
Figure 7. Effect of temperature.
570

Figure 8. Effect of reducing agent.
Figure 9. Reducing agents.
Figures 10 and 11. SEM image of Raney nickel before and after reaction.
Table 1. Recycling of Raney nickel.
Fresh Recycle 1 Recycle 2
Initial Final Initial Final Initial Final

conc conc %Dechlorination conc conc %Dechlorination conc conc %Dechlorination
25.2 6.68 73.55 54.383 18.8 65.43 77.46 29.598 61.79
571

3.10 Recycling of Raney nickel
The Raney nickel catalyst is checked for recyclability. The product formed is benzene. It
can poison the catalyst and recyclability is reduced. The second recycle itself reduced the
dechlorination by 8% and in a further recycle by more than 10%. This shows that the sites
are blocked by chloride ions, and agglomeration may occur. Recycling is possible because
contamination is less in the catalyst. This is evident from the SEM images of the Raney nickel
catalyst before and after dechlorination, as shown in Figure 10 and Figure 11.
3.11 Kinetics of Raney nickel

Hydrodechlorination reaction is conducted for 24 hours and 90.78% dechlorination is
achieved. Reaction follows first-order kinetics with a rate constant of 0.0136 min−1. To
enhance the reaction sodium borohydride is added as a reducing agent.
4 ANALYSIS OF PRODUCTS
Dechlorination of 1,4-dichlorobenzene using Raney nickel and palladium on activated car-

bon (0.5wt%) was done without reducing agents like hydrogen. It was observed that when
palladium is used as a hydrogenation catalyst, nearly 100% dechlorination is attained. For-
mation of benzene was confirmed by UV spectroscopy. Figure 12 shows a sharp peak visible
at 267 nm, which depicts the presence of a benzene ring.
Figure 12. Kinetics of Raney nickel.
Figure 13. UV spectroscopy.
572

Figure 14. HPLC peak for benzene.
Figure 15. HPLC analysis after reaction.
Benzene was separately injected into the HPLC column and the peak appeared at 4.5 min.
Afterwards, the reaction sample was analyzed in HPLC. The HPLC analysis after reaction result
is shown in Figure 15. A small peak appears at 4.457 min; it represents the presence of benzene.
5 CONCLUSIONS
Catalytic hydrodechlorination of 1,4-dichlorobenzene from wastewater was the aim of this

work, in which a hydrogenation catalyst like Raney nickel, palladium on activated carbon,
and bimetals were used. A hydrogenation catalyst such as Raney nickel and palladium showed
fast dechlorination compared to bimetallic catalysts. For Raney nickel, the hydrogenation
activity was comparable to that of palladium and the cost is also very much less. The Raney
nickel catalyst was further studied in detail by changing parameters like catalyst loading,
pH, initial concentration, the effect of salts, and the effect of temperature. Various reducing
agents were also tried for catalytic hydrodechlorination reaction. UV spectroscopy analysis
was done for various reaction mixture samples after the reaction, and the absorbance peak
matched to that of benzene.
573

REFERENCES
Cheng, R.O.N.G., Wang, J. & Zhang, W. (2007). Reductive dechlorination of p-chlorophenol by nano-
scale iron. Biomedical and Environmental Sciences, 20(5), 410–413.
Pearson, C.R. (1982). Halogenated aromatics. In O. Hutzinger (Ed.), Anthropogenic Compounds: Volume
3, Part B (The Handbook of Environmental Chemistry) (pp. 89116). New York: Springer-Verlag.
Sopoušek, J., Pinkas, J., Brož, P. Buršík, J., Vykoukal, V., Škoda, D, & Šimbera, J. (2014). Ag-Cu colloid
synthesis: Bimetallic nanoparticle characterisation and thermal treatment. Journal of Nanomaterials,
2014, 1.
Xia, C., Liu, Y., Xu, J., Yu, J., Qin, W. & Liang, X. (2009). Catalytic hydrodechlorination reactivity of
monochlorophenols in aqueous solutions over palladium/carbon catalyst. Catalysis Communications,
10(5), 456–458.
574

Controller tuning method for nonlinear conical tank system using

MATLAB/Simulink
S. Krishnapriya & R. Anjana

ABSTRACT: In the process industries the control of liquid level is mandatory. But the
control of nonlinear process is difficult. Many process industries use conical tanks because
their nonlinear shape provides better drainage for solid mixtures, slurries and viscous liquids.
Conical tanks are extensively used in the process industries, petrochemical industries, food
process industries and wastewater treatment industries. So, control of conical tank level is
a challenging task due to its nonlinearity and continually varying cross section. This is due
to the relationship between the controlled variable level and manipulated variable flow rate,
which have a square root relationship. The system identification of the nonlinear process is
made and studied using mathematical modeling with Taylor series expansion, and the real
time implementation is done in Simulink using MATLAB.
1 INTRODUCTION
Every industry faces the flow control and level control problem and have plentiful fea-
tures such as nonlinearity, time-delay, and time invariants. These features cause difficulties
in obtaining the exact model. Conical tanks are extensively used in the process indus-
tries, petrochemical industries, food process industries and wastewater treatment indus-
tries. The conical tank is generally nonlinear in nature due to its varying cross-sectional
area. Although many innovative methodologies have been devised in the past 50 years
to handle more complex control problems and to achieve better performances, the great
majority of industrial processes are still controlled by means of simple Proportional-Integral-
Derivative (PID) controllers. This seems to be because PID controllers, despite their simple
structure, assure acceptable performances for a wide range of industrial plants, and their
usage (the tuning of their parameters) is well known among industrial operators. Hence,
PID controllers are simple and easy if the process is linear. Since the process considered is a
nonlinear process, various other techniques are being implemented, which include Internal
Model Control (IMC) and fuzzy logic control. The IMC design procedure is exactly the
same as the open loop control design procedure. In addition, the IMC structure compen-
sates for disturbances and model uncertainty. The filter parameters in IMC are considered
and are used to tune the model of the given system to get the desired output. The use of
fuzzy logic controllers seems to be particularly appropriate, since it allows us to make use
of the operator’s experience and therefore to add some sort of intelligence to the automatic
control. Firstly, a PID controller has been designed by using the Ziegler-Nichols frequency
response method, and its performance has been observed. The Ziegler-Nichols tuned con-
troller parameters are fine-tuned to get satisfactory closed-loop performance. Secondly,
it has been proposed for the same system to use IMC and fuzzy logic controllers. A per-
formance comparison between the PID controller, IMC-based PID controller, and fuzzy
logic controller is presented using MATLAB/Simulink. Simulation results are studied, and
finally the conclusion is presented.
575

2 SYSTEM IDENTIFICATION
The system used is a conical tank, which is highly nonlinear due to the variation in area of
cross section. The controlling variable is inflow of the tank. The controlled variable is the
level of the conical tank. A level sensor is used to sense the level in the process tank and is
fed into the signal conditioning unit. The required signal is used for further processing. The
level process station is used to perform the experiments and to collect the data. One of the
computers is used as a controller. It consists of the software which is used to control the level
process station. The process consists of a process tank, reservoir tank, control valve, I to P
(I/P) converter, level sensor and pneumatic signals from the compressor.
When the setup is switched on, the level sensor senses the actual level. Initially the signal is
converted to a current signal in the range between 4 to 20 mA. This signal is then given to the
computer through a data acquisition cord. Based on the controller parameters and the set-point
value, the computer will take consequent control action and the signal is sent to the I/P converter.
Then the signal is converted to a pressure signal using the I/P converter. The pressure signal acts
on a control valve which controls the inlet flow of water into the tank. A capacitive type level
sensor is used to sense the level from the process and converts it into an electrical signal. Then the
electrical signal is fed to the I/V converter, which in turn produces a corresponding voltage signal
to the computer. The actual water level storage tank sensed by the level transmitter is fed back
to the level controller, and then compared with a desired level to produce the required control
action that will position the level control as needed to maintain the desired level. Now the con-
troller decides the control action. It is first given to the V/I converter and then to I/P converter.
The final control element (pneumatic control valve) is now controlled by the resulting air pres-
sure. This in turn controls the inflow to the conical tank and the level is maintained.
The system specifications (Rajesh et al., 2014) of the tank are as follows:
• Conical tank – Stainless steel body, height 70 cm, top diameter 35 cm, bottom diameter 2.5 cm
• Pump – Centrifugal 800 LPH
• Valve coefficient – K = 2
• Control Valve – Size ¼ pneumatic actuated type: Air to open, input 3–15 psi
• Rota meter range – 0–600 LPH.
2.1 Mathematical modeling

A mathematical model is a description of a system using mathematical concepts and language.
Generally modeling of linear systems involves direct derivations whereas nonlinear systems
require certain approximations to arrive at the solution (Vijayan & Avinashe, 2015). The Tay-
lor’s series method is simple and accurate over certain ranges near the steady state point. The
dynamic behavior of the liquid level h in the conical storage tank system is shown in Figure 2.
Figure 1. Level control of conical tank system.
576

Figure 2. Tank cross sections.
r R
tanθ = and also tanθ = (1)
h H
where R = Top radius of the tank

H = Total height of the tank
r = Radius at the liquid level (h)
h = Level of the liquid (variable)
By Newton’s law:
1 dh  
2
 R
Fin − Fout = *  A + 2π   * h2  (2)
3 dt   H

Output flow rate, Fout = K h (3)
dh
= α Fin h−2 − β h−3/2 (4)
dt
where (Vijula et al., 2014).
1
α = 2 (5)
 R
π
 H 
β = Kα (6)
dhs
At steady state, = αFis hs −2 − βhs −3/2 = 0 (7)
dt
Let y = (h − hs) and U = F − Fis (8)
dy  1
= −   βhs−5/ 2 y + α hs−2 U (9)
dt  2
 2  −5/2  dy   2α  −1/ 2
 β  hs  dt  + y =  β  hs U (10)
 dy 
τ + y = CU
 dt  (11)
Taking the Laplace transform:
Y (s) C
= (12)
U (s) τ s + 1
577

Table 1. Transfer function for the different height of the tank. [5].
Model Height Transfer function
1 10
y (s) 3.16
=
U ( s ) 62.08s + 1
2 15
y( s ) 3.87
=
U ( s ) 170.66 s + 1
Table 2. Process transfer function.
Model Height Transfer function
1 10
y( s ) 0.03097
=
U ( s ) 9.312s 2 + 28.15s + .451
2 15
y( s ) 0.03793
=
U ( s ) 36.942s + 12.16s + 0.451
2
where
 2
τ =   hs−5/ 2 (13)
 β
2α −1/2
C= hs (14)
β
The transfer function of the sensor is:

0.0098
G v (s) = (15)
0.15s + 0.451
The modeling of transfer function for the conical tank with height 70 cm, top diameter 35
cm and height of liquid level at 15 cm is obtained as:
y( s ) 0.038
= (16)
U ( s ) 36.942s 2 +12.16s + 0.451
3 CONTROLLER TUNING METHODS
3.1 PID Controller

The PID control is simple in principle, easy to tune, robust and a successful realistic application,
which is still widely used in industrial process control. The PID controller is a fundamental part
of the control loop in the process industry. Even though many advanced control modes are
based on a PID control algorithm, the conventional PID control algorithm cannot achieve ideal
control effect in any practical production process with nonlinear and time varying uncertainty.
de(t ) t
u(t ) = K pe(t ) + K d K i ∫ e(τ )dτ (17)
dt 0
578

3.2 IMC-BASED pid controller
The Internal Model Control (IMC) is commonly used to provide a transparent mode for
the design and tuning of various types of control. The IMC-based PID tuning method is a
clear trade-off between closed-loop performance and robustness to model inaccuracies, and
the tuning is achieved with a single tuning parameter. Also, it allows good set-point track-
ing but sulky disturbance response, especially for the process with a small time-delay/time-
constant ratio. However, for many process control applications, disturbance rejection for the
unstable processes is much more important than set-point tracking. Hence, controller design
that emphasizes disturbance rejection rather than set-point tracking is an important design
problem that must be taken into consideration. In process control applications, model-based
control systems are often used to track set-points and reject low disturbances. The IMC
design procedure is the same as the open loop control design procedure. Unlike open loop
control, the IMC structure compensates for disturbances and model uncertainties. The IMC
filter tuning parameter ‘λ’ is used to avoid the effect of model uncertainty.
The various steps in the IMC system design procedure (Vijayan & Avinashe, 2015). are:
Step 1: Factorization: It includes factorizing the transfer function into invertible and non-
invertible parts.
Step 2: Form the idealized IMC controller. The ideal internal model controller is the inverse
of the invertible portion of the process model.
Step 3: Adding a filter. Now a filter is added to make the controller stable.
IMC (Fathima et al., 2015) is compared with conventional PID, giving:
θ θ
τ+ θ +τ
Kc = 2 Ti = + τ Td = 2
θ (18)
 2 θ 
K  λ+  2 + τ 
 2  2 
IMC filter tuning parameter, λ = 1.
3.3 Fuzzy logic controller

The ideas of fuzzy set and fuzzy control were introduced by Zadeh (1996) to control systems
that are structurally difficult to model. Mamdani was the first person to use fuzzy logic for con-
trol purposes (Lee 1990). Fuzzy systems can transform vague information and expert knowl-
edge into computable numerical data. It incorporates heuristics, developed by experts and
Figure 3. The IMC structure.
Figure 4. The equivalent feedback form to IMC.
579

Table 3. Comparison of the PID and IMC-based PID parameter values.
Kp Ti Td
PID 57.16 10.78 22.26

IMC-based PID 50.21 1.16 0.36
Figure 5. Simulink diagram and response of an IMC-based PID and a PID.
Figure 6. Simulink diagram and response of PID and fuzzy controller.
580

Table 4. Performance analysis.
Rise time (sec) Setting time (sec)
IMC based PID 5.4766 65.9554

PID 9.8523 83.2704
ISE IAE ITAE
PID 26.57 55.3 4000

IMC based PID 16.81 41.99 2763
Fuzzy 14.93 27.56 967.9
operators into automatic control. MATLAB/Simulink provides tools to create and edit fuzzy
inference systems within the framework. It is also possible to integrate the fuzzy systems into
simulations with Simulink. These are: fuzzy linguistic variable representing the level error (e);
change of level error (de); and, the output control effort (u), respectively. A Fuzzy Logic System
(FLS) can be defined as the nonlinear mapping of an input data set to a scalar output data.
Firstly, a crisp set of input data is gathered and converted to a fuzzy set using fuzzy linguis-
tic variables, fuzzy linguistic terms and membership functions. This step is known as fuzzi-
fication. Afterwards, an inference is made based on a set of rules. Lastly, the resulting fuzzy
output is mapped to a crisp output using the membership functions, in the defuzzification
step. Mainly Fuzzy Logic Controllers (FLC) are implemented on nonlinear systems which
yield better results. In designing the controller, the number of parameters (Ilyas et al., 2013)
needs to be selected, and then the membership function and rules are selected based on heu-
ristic knowledge.
The response of IMC and fuzzy controllers are compared with convectional a PID
controller.
The performance is studied by evaluating Rise Time, Settling Time, Integral Square Error,
Integral Absolute Error and Integral Time Absolute Error.
5 CONCLUSION
The controlling of nonlinear process is a challenging task and nonlinearity of the conical
tank is analyzed. Modeling of transfer function of the system is done by using system iden-
tification. Various controllers are simulated in MATLAB/Simulink. An open loop step test
method is used to find the proportional gain, delay time and dead time. Here Taylor series
approximation is used for the nonlinear approximation, because of its accuracy compared to
other nonlinear approximation techniques. The simulation results show that the IMC-based
PID controllers have minimum settling time and rise time in order to reach steady state value,
as compared to the conventional controller. After analyzing simulated response of models,
the fuzzy controller is found to be the more excellent controller than the IMC-based PID and
PID controllers.
REFERENCES
Fathima, M.S., Banu, A.N. Nisha, A. & Ramachandran, S. (2015). Comparison of controllers for a flow
process in a conical tank. International Journal, 1, 145–148.
Ilyas, A., Jahan, S. & Ayyub, M. (2013). Tuning of conventional PID and fuzzy logic controller using
different defuzzification techniques. International Journal of Scientific & Technology Research, 2(1),
138–142.
581

Lee, C.C. (1990). Fuzzy logic in control systems: fuzzy logic controller. I. IEEE Transactions on systems,
man, and cybernetics, 20(2), 404–418.
Rajesh, T., Arun., S. & Siddharth, S.G. (2014). Design and implementation of IMC based PID control-
ler for conical tank level control process. International Journal of Innovative Research in Electrical,
Electronics, Instrumentation and Control Engineering, 2(9), 2041–2045.
Sharma, A. & Venkatesan, N. (2013). Comparing PI controller performance for non-linear process
model. International Journal of Engineering Trends and Technology, 4(3), 242–245.
Vijayan, S. & Avinashe, K.K. (2015). IMC based PID and fuzzy controller for nonlinear conical tank level
process. IJEECS ISSN 2348–117X Volume 4, Special Issue September 2015.
Vijula, D.A., Vivetha, K., Gandhimathi, K. & Praveena, T. (2014). Model based controller design for
conical tank system. International Journal of Computer Applications, 85(12), 8–11.
Zadeh, L.A. (1996). Fuzzy logic = computing with words. IEEE transactions on fuzzy systems, 4(2),
103–111.
582

Study on multi walled carbon nanotubes synthesis, modeling

and applications
K. Krishnarchana & N. Manoj

Sushree Sangita Dash

Propellant Fuel Complex, VSSC, ISRO, Thiruvananthapuram, India
ABSTRACT: This study investigates the synthesis of multi walled nanotubes by the decom-
position of acetylene over misch metal catalyst in a chemical vapour deposition reactor. The
synthesized CNTs were analyzed by different spectroscopic techniques. Also, a kinetic model
for MWNT growth is proposed to investigate the dependence of the flow rate of precursor
on CNT production rate. The model is validated by comparing its predictions with a set of
experimental measurements and is simulated in MATLAB software. The experimental results
were found to agree well with the theoretical predictions obtained from the model. In addi-
tion to the synthesis and modeling of CNTs, this work also embodies a technique for conduc-
tive coating using multi walled carbon nanotubes.
1 INTRODUCTION
Carbon nanotubes attracted a lot of researchers from academia to industry because of their
remarkable mechanical and electronic properties, viz. high thermal and electrical conductiv-
ity, high aspect ratio, high tensile strength and low density, when compared with conventional
materials. And also finds promising applications in many fields such as field and light emis-
sion, biomedical systems, nanoelectronic devices, nanoprobes, nanosensors, Conductive com-
posites and energy storage. The CNTs are either single walled carbon nanotubes (SWNTs)
or multi walled carbon nanotubes (MWNTs). The biggest challenge in developing potential
applications for CNTs lies in the production of pure CNTs at affordable prices. The commonly
used CNT synthesis techniques are Arc discharge, Laser ablation, Chemical vapor deposition
(CVD), Electrolysis, Flame synthesis, etc. Among these methods, Chemical vapor deposition
is considered as cheap, simple and most promising way for large-scale synthesis of CNTs [1–3].
The CNT deposition profiles inside a CVD reactor strongly depend on various parameters
such as reaction temperatures, feed gas flow rates, carrier gas flow rates, catalyst type, etc. These
reaction conditions can vary throughout the reactor, affecting the yield as well as rate of the
reaction. Therefore it is very important to develop a model of the system as an important aid
in studying the CNT growth process since it can envisage the yield without undertaking expen-
sive experimental studies. Also helps in optimizing the process thereby enhancing CNT scale
up process. This work is solely dedicated to the formation of MWNTs via the decomposition
of acetylene over misch metal catalyst and its characterization. Also, aims to the modeling of
CNT synthesis process and finally the development of conductive coating using MWNTs.
2 SYNTHESIS AND CHARACTERIZATION OF MWNTS
Multi walled nanotubes are synthesized by the decomposition of acetylene over an alloy
of misch metal catalyst powder in a CVD reactor. The CVD reactor consist of a tubular
583

furnace with a quartz tube (55 mm diameter and 1.5 m long for small scale production)
and feed inlets for C2H2, CH4, H2 and Ar. The quartz tube has a heated zone, which acts as
the reaction chamber. The furnace temperature was maintained at 750°C. The flow rates
of the various gases were monitored and controlled using flow meters and solenoid valves.
The process is carried out in presence of catalyst and the carrier gas. The quartz boat was
cleaned with acetone, dried before the process and catalyst as taken in the quartz boat for
the synthesis of CNTs. The synthesis of CNTs in CVD was done by using LABVIEW
software.
A particular amount of the catalyst is taken in a quartz boat and is inserted in the center
of a quartz tube. Then the furnace is heated to 500°C in argon atmosphere. The flow rate of
argon is maintained at 160 sccm for 40 minutes. When the temperature reaches 500°C, hydro-
gen is introduced into the quartz tube for about 30 minutes, in order to remove any oxygen on
the surface of the alloy hydride catalyst. The hydrogen flow is then stopped and the furnace is
heated up to the desired reaction temperature of 750°C and this is followed by the introduc-
tion of acetylene at a flow rate of 70 sccm. The reaction is carried out for 90 minutes and
thereafter the furnace is cooled to room temperature. Argon flow is maintained throughout
the experiment. The weight of as grown CNT was taken. The procedure is repeated for vari-
ous flow rate of acetylene.
The synthesized CNTs may contain impurities such as graphite nanoparticles, amorphous
carbon, smaller fullerenes, and metal catalyst particles. These impurities were separated from
the carbon nanotubes by following purification steps; Air Oxidation (Removes carbonaceous
impurities such as amorphous carbon and helps expose the catalytic metal surface enclosed
in the carbon nanotube for further purification techniques), Acid Treatment (Removes metal
catalyst and small fullerene isomers from CNTs and, also helps in its functionalization), Wash-
ing (Based on size or particle separation that separates CNTs from metal nanoparticles, poly-
aromatic carbons and fullerenes) and Drying (Remove any volatile products, if present). After
purification, the CNTs are characterized by Raman spectroscopy, HRTEM, SEM, TGA and
FTIR.
3 KINETIC MODELING
Catalytic graphitisation of carbon was used to explain the synthesis of multi walled carbon
nanotubes from acetylene using the catalytic chemical vapour deposition method.
Catalytic graphitisation involves carbon dissolution, adsorption and reaction to produce
CNTs. Equation (1) presents the mechanisms of catalytic graphitisation of acetylene to CNTs
using the CVD technique, where acetylene and possibly the cracked fractions under heat are dis-
sociated into carbon atoms. The carbon atoms are deposited and adsorbed on the catalyst sur-
face, which in turn reacted with each other to form C-C bonds to produce the carbon nanotubes.
Mm
2C 2H2 4C + 2H2
→0
750 C
(1)
nC ↔ C n − Mm → m [ CNTs ] + Mm
The rate of catalytic graphitization,
rcc = fn (Dissolution, Adsorption, chemical reaction) (2)
At temperature less than 750°C, no chemical reaction or no CNTs production was observed,
but as the temperature reaches 750°C, production of CNTs was occur which indicating the
complete decomposition of acetylene. Therefore, at this temperature, dissolution process was
assumed to become non rate limiting, hence the rate of catalytic graphitisation formation and
acetylene consumption becomes equal.
584

rcc = − rc2H = fn ( Adsorption, chemical reaction) (3)
2
The catalytic graphitization of acetylene to CNT is normally the case with solid catalyzed
reaction that can be expressed by the rates of reaction catalyzed by solid surfaces per unit
mass as,
1 dN C
− rc2 H = = kθ c n (4)
2
WMm dt
Here, Langmuir Hinshelwood mechanism is adopted to obtain the reaction rate and equi-
librium constants.
kKC n
−rc2 H = (5)
2
1 + KC n
1 1 1
= n
+ (6)
rc2 H kKC k
2
Equation (6) is used to determine the kinetics parameters used in computing the model.
The model which represents the production CNT by CVD is obtained as,
kKCAn exp (1 − θ )
r= C (7)
1+ K k1t
4 DEVELOPMENT OF CONDUCTIVE COATING
The high electrical conductivity and low density of CNTs makes them a suitable material
for coating applications. Here, the coating is to be developed particularly for cryogenic tank
exteriors. Currently a PU base gray conductive coating is used for the purpose which is of
higher density. CNTs, because of its innate low density are expected to perform much better
than this conventional strategy.
The experimental procedure for the development of conductive coating involves two steps;
Substrate preparation and CNT dispersion in solvents. In Substrate preparation, PU foam
of 10 × 10 × 2 cm was taken. Two coatings were applied over the foam; first one being the PU
coating as VBC to prevent any moisture permeation into the foam from the outside atmos-
phere and second one, the CNT conductive coating. Electrically conductive coatings were
prepared by dispersing CNT in solvent. Before applying the second coating, the CNTs were
totally dispersed in either acetone or toluene by sonication for 3 hours. The dispersed solu-
tion is then applied on the substrate, weighed and dried for 15 minutes to obtain a uniformly
coated conductive layer on the PU substrate. The conductivity of the coating was measured
by means of surface resistivity meter. The procedure was repeated until the expected conduc-
tivity (Conductivity of MWNT, 10−3 S/m) was obtained.
5.1 Synthesis and characterization

The MWNT yielded by the chemical vapor deposition of acetylene gas over misch metal
catalyst was 87.06%. Fig. 1 shows the SEM image of the surface morphology of MWNTs.
The SEM image discloses that the CNTs are less straight and length in several micrometers.
The white region indicates the aggregated CNTs and the regions are magnified in order to
study the morphology.
585

Figure 2. HRTEM image of MWNT.
Figure 1. SEM image of MWNTs.
Figure 3. Raman spectra of MWNT. Figure 4. TGA of MWNT.
Figure 2 is the HRTEM image of MWNTs which shows that MWNTs have a hollow
structure. And image reveals the diameter of the CNTs. It shows that MWNT has an inner
diameter of 12.5 nm and outer diameter of 38 nm.
Fig. 3 shows the Raman spectra of MWNTs, from which Multiwall structure of CNT
was identified. The peaks present at 1577.96 cm-1, 1347.1 cm-1 and 2693.3 cm-1 represents
the G, D and G’ modes in Raman spectra. The presence of CNT is identified by a G line at
1577.96 cm-1. The RBM mode was not there. Hence, the multi walled structure of CNT was
confirmed. The D band corresponds to 1347 cm-1 is related defects of graphitic sheets or
carbonaceous particle at the surface of the tube.
Figure 4 is a plot for the weight loss in % vs. the oxidation temperature, measured by heat-
ing up the MWNTs in a TGA. The weight loss curve between 100 and 800°C was plotted by
adjusting about 100% for the weight loss at 800°C, in which the actual weight was presumably
the weight of catalyst (usually 10% of total weight). Weight losses below 200°C and From
the TGA plot, it can be seen that there was no significant weight loss up to 300°C. After this
temperature, a slight decrease in the weight can see due to the burning of amorphous carbon
until the temperature reaches 500°C. While reaching 500°C, a sharp decrease in the weight
can be seen due to the burning of MWNTs. On reaching 650°C, all the MWNTs have burned.
There was no residual weight percentage at 650°C which implies that the MWNTs produced
at 750°C were 100% pure.
Fig. 4 shows the FTIR image of MWNTs synthesized over misch metal catalyst at 750°C.
The wave number 3410 cm-1, 1726 cm-1 and 1594 cm-1 represents the O-H stretches of the
terminal carboxyl group, the carboxyl C = O groups and the C = C stretching respectively.
From this it was clear that by the acid treatment of MWNTs improves its interfacial interac-
tion. Hence it can be used to make matrix structures.
5.2 Simulation results

The effects of varying acetylene concentration on CNTs production rate were investigated at
the range of 50 ml/min – 90 ml/min at 750°C, while 160 ml/min Ar flow was used as the car-
586

Figure 5. FTIR of MWNT.
rier gas flow and the results obtained are presented in Figure 6(a). A kinetic model equation
was developed to predict the production rate of MWNTs as
0.775 × 2.206 × 10−15 exp( 1− 1)

r= −15
CC 2 H 2 4 (8)
1+ 2.206 × 10 5400 × 2.86 × 10−15
The modeled equation predicts the rate of production of CNTs at various acetylene con-
centrations. The results obtained were shown in Table 1.
It is evident that the plots are comparable, even though they are not exact similar. This may
due to the assumption that the carbon atoms are occupied on the entire surface of the catalyst.
Hence, the value of fraction of surface area occupied by the carbon atom (θ) is taken as 1.
It is expected that a more accurate value of θ may yield better results.
5.3 Conductive coating

Assessment of MWNTs as conductive coating material investigated through several experi-
ments. The best results obtained were shown in Table 2.
Trials 1,2,3 and 4 showed promising results, which make CNT based conductive coating a
suitable candidate for cryogenic tank insulation. Especially, the trial 15 evolved a light weight
system which improves the ease of application with enhanced performance. The optimization of
the process was also undertaken which enhances the repeatability and reliability of the technique.
6 CONCLUSION
Chemical vapor deposition based production of carbon nanotubes yields good quality, uni-
formly and well aligned nanostructures. In this work multi walled nanotubes (87.06%) are
synthesized by chemical vapor deposition of acetylene over misch metal catalyst at 750°C.
The CNTs were characterized by SEM, HRTEM, Raman spectroscopy, TGA and FTIR.
From Raman spectra and HRTEM, the presence of CNTs was identified and confirmed.
FTIR analysis revealed that the CNTs are functionalized during acid treatment. SEM image
gives an idea about the morphology and structure of MWNTs. The TGA results show that
the CNT synthesized at 7500C was almost 100% pure.
In the second section of the work, a kinetic model was developed, for the MWNT synthe-
sis by CVD, to study the effect of flow rate of acetylene on the CNT production rate. The
model equation was based on the experimental data. The theoretical prediction from the
model equation and experimental data are comparable. Here the maximum yield obtained is
0.374 mg/sec at acetylene concentration of 4287.32 ppm.
Synthesized MWNTs are used to develop a CNT based conductive coating (conductivity:
10–4 S/m) which is a suitable coating for cryogenic tank insulation. The obtained coating
587

Table 1. Experimental and computed rate of production of CNTs at varying con-
centration of acetylene.
CNT production rate, r (mg/sec)

Concentration of
acetylene, CA (ppm) Experimental value Computed value
2836.48 0.098 0.071

3248.34 0.114 0.123
3439.72 0.135 0.155
3624.75 0.214 0.191
3968.74 0.351 0.274
4287.32 0.37 0.374
Table 2: Results of trials conducted on conductive coating.
No. of trials Materials used Sonication Results
1 CNT: 1.5 g 2 hr Conductive, 10-5 S/m

Acetone: 50 ml (spray coating)
But, CNTs are coming out
2 CNT: 0.03 g 3 hrs Conductive, 10-4 or 10-5 S/m
Acetone: 5 ml
PU: 3.21 g
3 CNT: 0.03 g 3 hrs Conductive, 10-4 S/m
Acetone: 10 ml
PU: 5.58 g
4 CNT: 0.05 g 3 ½ hrs Conductive, 10-4 S/m
Acetone: 20 ml
PU: 0.14 g
Figure 6. (a) Effect of flow rate of acetylene on CNTs production rate (b) Computed results of CNTs
production rate.
Figure 7. Conductive coating (Conductivity: 10–4 or 10–5 S/m, Area: 0.01 m2).
588

is comparatively light in weight which improves the easy of application with enhanced per-
formance. The optimization of the process was also undertaken which enhances the repeat-
ability and reliability of the technique.
ACKNOWLEDGEMENT
The satisfaction and euphoria on the successful completion of any task would be incom-
plete without mentioning the people who made it possible whose constant guidance and
encouragement crowned out effort with success. I express my heartfelt thanks to Sushree-
sangita dash (External guide), Manoj N (Internal guide), V.O. Rejini (HOD), S.K. Manu
(Dy. Manager, PFC, VSSC), Sriram P. (Engineer SC), all other staffs of VSSC, all the fac-
ulty members of the department of Chemical Engineering, my friends and my family.
NOTATIONS
Ka: Rate constants for adsorption

Kd: Rate constant for desorption
K: Rate constant for chemical reaction.
WMm: Weight of catalyst used
Nc: No. of moles of carbon
N: Order of reaction
Θ: Fraction of Mm surface occupied by carbon atoms
CA: Concentration of reactant
K1: Rate constant, proportional to the diffusion coefficient of carbon
T: Time
REFERENCES
[1] Andrea Szabó, CaterinaPerri, Anita Csató, Girolamo Giordano, DaniloVuono and János B. Nagy,
“Synthesis Methods of Carbon Nanotubes and Related Materials”, Materials 2010, 3, 3092–3140.
[2] KalpanaAwasthi, Anchal Srivastava and O.N. Srivastava, “Synthesis of carbon nanotubes ”, Physics
Department, Banaras Hindu University, Varanasi-221 005, India.
[3] Adedeji E. AgboolaRalph W. Pike T.A. Hertwig Helen H. Lou, “Conceptual design of carbon nano-
tube processes”, Clean Techn Environ Policy (2007) 9:289–311.
[4] KochandraRaji and Choondal B. Sobhan, “Simulation and modeling of carbon nanotube synthesis:
current trends and investigations”, Nanotechnolgy Rev 2013; 2(1): 73–105.
[5] Sunny EsayegbemuIyuke, SakaAmbaliAbdulkareem, Samuel Ayo Afolabi, and Christo H.
vZPiennar, “Catalytic Production of Carbon Nanotubes in a Swirled Fluid Chemical Vapour Depo-
sition Reactor”, International Journal of Chemical Reactor Engineering, Volume 5, 2007, Note S5.
[6] K. Raji, Shijo Thomas, C.B. Sobhan, “A chemical kinetic model for chemical vapor deposition of
carbon nanotubes”, Applied Surface Science 257 (2011) 10562–10570.
[7] O. Levenspiel, “Chemical Reaction Engineering” third ed., Wiley India Pvt. Ltd., 2006.
[8] M.N. Masri, Z. M. Yunus, A.R.M Warikhand A.A Mohamad (2010), “Electrical conductivity and
corrosion protection properties of conductive paint coatings”, anticorrosion methods and materials,
vol 57, issue 4, pp. 204–208 (2010).
589

Removal of nickel from aqueous solution using sargassum

tenerrimum powder (brown algae) by biosorption: Equilibrium,
kinetics and thermodynamic studies
M. Tukaram Bai, P. Venkateswarlu & Y.V. Anudeep

Department of Chemical Engineering, Andhra University, Visakhapatnam, India
ABSTRACT: In the present study the biosorption of nickel onto sargassum tenerrimum pow-
der (brown algae) from an aqueous solution was studied. The equilibrium study was carried
out for parameters: agitation time (1–210 min) (t), biosorbent size (45–300 µm) (dp), biosorbent
dosage (2–24 g/L) (w), pH of aqueous solution (1–8), initial concentration of nickel in aqueous
solution (5–150 mg/L) (C0), and temperature (283–323°K) of aqueous solution on biosorption
of the metal (nickel) were studied. In the present investigation the equilibrium data was well
explained by Langmuir, Temkin, and Redlich and Peterson with a correlation coefficient of
0.99, and followed by a Freundlich isotherm. The kinetic studies reveal that the biosorption
system obeyed the pseudo second order kinetic model by considering the correlation coefficient
value as 0.99. From the values of ∆S, ∆H and ∆G it is observed that the biosorption of nickel
onto sargassum tenerrimum powder was irreversible, endothermic and spontaneous.
Keywords: Nickel, biosorption, sargassum tenerrimum, algae, isotherms, kinetics,

thermodynamics
1 INTRODUCTION
All living organisms require heavy metals in low concentrations, but high concentrations of
heavy metals are toxic and can cause cancer (Koedrith et al., 2013). Nowadays the environ-
ment is threatened by an increase in heavy metals. Therefore, in recent years the removal of
heavy metals has become an important issue (Nourbakhsh et al., 2002). Methods for remov-
ing metal ions from aqueous solution mainly consist of physical, chemical and biological
technologies. Conventional methods for the removal of heavy metal ions from wastewater,
such as chemical precipitation, flocculation, membrane filtration, ion exchange, and electro-
dialysis electrolysis, are often costly or ineffective for the treatment of low concentrations of
pollutants (Wang & Chen, 2009). Biological uptake is a promising approach that has been
studied in the past decade. This process is a good candidate for replacing old methods (Pinto
et al., 2011). High efficiency, removal of all metals even at low concentrations, being economi-
cal, and energy independence are the main advantages of biological uptake which present
this process as being a viable new technology (Bai & Abraham, 2002). Biosorption is used to
describe the passive non-metabolically mediated process of metal binding to living or dead
biomass (Rangsayatorn et al., 2002). Water pollution by heavy metals is globally recognized
as being an increasing environmental problem since the start of the Industrial Revolution
in the 18th century (Dàvila-Guzmàn et al., 2011). Heavy metals may come from different
sources such as electroplating, textile, smelting, mining, glass and ceramic industries as well
as storage batteries, metal finishing, petroleum, fertilizer, pulp and paper industries. Nickel
is one of the industrial pollutants, possibly entering into the ecosystem through soil, air, and
water. Nickel is a toxic heavy metal found in the environment, as a result of various natu-
ral and industrial activities. The higher concentration of the nickel causes poisoning effects
like headache, dizziness, nausea, tightness of the chest, dry cough and extreme weakness
591

(Krishna & Swamy, 2011). So, it is very much essential to remove nickel from wastewater.
The objective of the present work is to explore the potential of a biosorption technique for
the removal of nickel from aqueous solutions using cheap and abundantly available materials
like sargassum tenerrimum powder. The experiments are undergone in batch process for the
equilibrium studies, kinetics and thermodynamics of biosorption.
2.1 Preparation of the biosorbent

Fresh samples of sargassum tenerrimum, a species of the brown algae, were collected from
Rushikonda beach in Visakhapatnam, Andhra Pradesh. The brown algae were washed thor-
oughly with distilled water ten times to remove sand and dirt completely. The cleaned brown
alga was dried in sunlight, until the moisture was completely removed. The dried biomass was
ground to powder by using a mechanical grinder. The ground powder was then sieved and
separated into various sizes (45, 53, 75, 150 and 300 µm) by using British Standard Sieves.
The powders of these size fractions were then stored in separate plastic airtight containers at
room temperature and used as biosorbent for subsequent analysis.
2.2 Preparation of aqueous nickel stock solution

The quantity of 4.1582 g of 97% pure of NiCl2.6H2O was dissolved in 1 L of distilled water
to prepare 1,000 mg/L of nickel solution. 100 mg/L nickel solution was prepared by diluting
100 mL of 1,000 mg/L nickel stock solution with distilled water in a 1,000 mL volumetric
flask up to the mark. The pH of the aqueous solution was adjusted to the desired value by
the addition of 0.1 N HCl or 0.1 N NaOH solutions.
2.3 Procedure
The procedures adopted to evaluate the effects of various parameters viz. agitation time (t),
biosorbent size (dp), biosorbent dosage (w), pH of aqueous solution, initial concentration of
nickel in aqueous solution (C0), and temperature of aqueous solution, on the biosorption of
metal (nickel) are explained below.
50 mL of aqueous solution containing 20 mg/L of initial concentration of nickel was
taken in a 250 mL conical flask. 10 g/L of 45 µm size biosorbent was added to the flask.
The conical flask was then kept on an orbital shaker at room temperature (30°C) and was
shaken for one min. Similarly, 21 more samples were prepared in conical flasks by adding
10 g/L of biosorbent and agitating for different time periods from 2 to 210 min. For the
resulting agitation equilibrium time of 120 min the further experiments were repeated for
varying biosorbent sizes viz. 75, 150 and 300 µm. The resulting optimum biosorbent size
was 45 µm. The above procedure was repeated for different adsorbent dosages of 4, 6, 8,
10, 12, 14, 16, 18, 20, 22 and 24 g/L at equilibrium agitation time (120 min) and biosorbent
size is optimum (45 µm). The equilibrium biosorbent dosage was found to be 18 g/L. To
determine the effect of pH on nickel biosorption, 50 mL of aqueous solution was taken
in each of 12 conical flasks. The pH values of aqueous solutions were adjusted to 1, 2, 3,
3.5, 4, 4.5, 5, 5.5, 6, 6.5, 7 and 8 in separate 250 mL conical flasks. 18 g/L of 45 µm size
biosorbent was added to each of the conical flasks. The influence of initial concentration
on biosorption of nickel was determined as follows: 50 mL of aqueous solutions, each of
different nickel concentrations of 5, 10, 20, 30, 40, 50, 60, 75, 100, 125 and 150 mg/L were
taken in 11 250 mL conical flasks. 18 g/L of 45 µm size biosorbent was added to each of
the conical flasks. The flasks were agitated on an orbital shaker for equilibrium agitation
time at room temperature. The samples were allowed to settle and then filtered separately.
The samples thus obtained were analyzed in atomic absorption spectroscopy (AAS) for the
final concentrations of nickel in aqueous solutions.
592

3.1 Effect of agitation time

The effect of agitation time on percentage biosorption is shown in Figure 1. For 45 µm
biosorbent size, the % biosorption is significant in the first 1 min up to 81.25%. The %
biosorption is increased gradually from 81.25% to 93% between the time intervals of 1 to
120 min. After the agitation time of 120 min, % biosorption remains constant, indicating the
attainment of equilibrium conditions. The same equilibrium agitation time of 120 min was
reported for the biosorption of nickel by brown alga fucus vesiculosus (Mata et al., 2008) and
immobilized algal cells (Al-Rub et al., 2004).
3.2 Influence of biosorbent size

The experiments are carried out for various biosorbent sizes – 45, 75, 150 and 300 µm—keep-
ing other parameters constant. Figure 2 shows the graph drawn between % biosorption of
nickel against biosorbent size. The biosorption of nickel decreases from 93% (1.86 mg/g) to
Figure 1. Effect of contact time % on removal of nickel metal.
Figure 2. Effect of biosorbent size on % removal of nickel.
593

79% (1.58 mg/g) as the biosorbent size increases from 45 to 300 µm. With a decrease in bio-
sorbent particle size, the surface area of the biosorbent increases and the number of active
sites available on the biosorbent are better exposed to the biosorbate.
3.3 Effect of biosorbent dosage

To study the effect of biosorbent dosage on % biosorption of nickel, the biosorbent dosage
is varied from 2 to 24 g/L (2, 4, 6, 8, 10, 12, 14, 16, 18, 20, 22 and 24 g/L). A plot is drawn
between % biosorption of nickel and biosorbent dosage in Figure 3. The biosorption of
nickel is increased from 90.5% (9.05 mg/g) to 96.35% (1.07 mg/g) with an increase in biosorb-
ent dosage from 2 to 18 g/L. Such behavior is obvious because with an increase in biosorb-
ent dosage, the number of active sites available for nickel biosorption would be more. The
change in percentage biosorption of nickel is minimal, from 96.35% (1.07 mg/g) to 96.58%
(0.804 mg/g) when ‘w’ is increased from 18 to 24 g/L. Hence, optimum biosorbent dosage of
18 g/L is considered to study all other parameters.
3.4 Effect of pH
The pH of aqueous solution is drawn against % biosorption of nickel in Figure 4. The %
biosorption of nickel is increased from 89.65% (0.996 mg/g) to 97% (1.077 mg/g) as pH is
increased from 1 to 4.5 and decreased beyond pH value of 4.5. In the case of lower pH val-
ues, the occupation of the negative sites of the biosorbent by H+ ions leads to a reduction
of vacancies for nickel ion and consequently causes a decrease in nickel ion biosorption.
As the pH is raised, the ability of the nickel ions to compete with H+ ions also increases.
Although the sorption of nickel ions is raised by a growing pH, a further increment of pH
causes a decline in biosorption due to the precipitation of nickel hydroxides. The predomi-
nant adsorbing forms of nickel are nickel and NiOH+, which occur in the pH range of 4–6.
The value of pH at 4.5 is considered as being optimum for the study of other parameters. The
functional groups like aliphatic C-H, SO3 stretching, C-O and C = O stretching, aromatic—
CH stretching and amine groups of the biosorbent were responsible for nickel biosorption.
Similar results were reported for biosorption of nickel by waste pomace of olive oil factory
(Nuhoglu & Malkoc, 2009). The optimum pH ranging from 4 to 5 was reported by Aksu
et al. (2006), Özer et al. (2008) and Congeevaram et al. (2007) for the biosorption of nickel by
using various biosorbents like dried chlorella vulgaris, enteromorpha prolifera and aspergillus
species respectively.
Figure 3. Effect of biosorbent dosage on removal of % nickel.
594

Figure 4. Effect of pH on removal of % nickel.
Figure 5. Effect of pH on removal of % nickel.
Figure 6. Effect of temperature on % removal of nickel.
595

3.5 Effect of initial concentration of nickel
Figure 5 shows the graph between initial concentration of aqueous solution (C0) and %
iosorption of nickel at w = 18 g/L and pH = 4.5. The effect of initial nickel ion concentration
b
is investigated in the range of 5–150 mg/L. The percentage biosorption of nickel is decreased
from 98.85% to 53.75% with an increase in C0 from 5 to 150 mg/L while the uptake capac-
ity is increased from 0.274 to 4.419 mg/g respectively. Such behavior can be attributed the
increase in the amount of biosorbate to the unchanging number of available active sites on
the biosorbent. Similar results were reported by Pahlavanzadeh et al. (2010) and Gupta et al.
(2010), using brown alga and treated alga (oedogonium hatei) for the biosorption of nickel.
3.6 Effect of temperature

The effect of temperature on % biosorption of nickel is shown in Figure 6. The study was
further extended at initial concentrations of 5, 10 and 20 mg/L. For the initial concentra-
tion (C0) of 5 mg/L, the % biosorption is increased from 96.25% to 99% as the temperature
increases from 283°K to 333°K.
4 ADSORPTION ISOTHERMS
4.1 Freundlich isotherm

Freundlich (1907) presented an empirical adsorption isotherm equation that can be applied
in cases of low and intermediate concentration ranges.
Taking logarithms on both sides, we get:
log qe = log Kf + n log Ce (1)
A Freundlich isotherm is drawn between log qe and log Ce, in Figure 7 for the present data.
The equation obtained is:
log qe = 0.381log Ce + 0159 (2)
with a correlation coefficient of 0.97. The Freundlich constant (Kf) is found to be 1.054 and
the n-value of 0.381 lies between 0 and 1, indicating the applicability of the Freundlich iso-
therm to the experimental data.
Figure 7. Freundlich isotherm for biosorption of nickel.
596

4.2 Langmuir isotherm
Irving Langmuir developed an isotherm named as the Langmuir isotherm (Langmuir, 1918).
It is the most widely used simple two-parameter equation. The Langmuir relationship is
hyperbolic and is:
 qe 
 qm  = bCe / (1 + bCe ) (3)
where
Ce is the equilibrium concentration (mg/L),
Qe is the amount of nickel adsorbed (mg/g).
Equation 3 is rearranged as:
 Ce  1
 qe  = bqm + (1/ qm )Ce (4)
The Langmuir isotherm (Figure 8) for the present data is represented as:
 Ce 
 qe  = 0219Ce + 0.920 (5)
with a good linearity (correlation coefficient, R2 = 0.98) indicating a strong binding of nickel
ions to the surface of sargassum tenerrimum powder. The qm and b values are 4.566 mg/g and
0.98 respectively. The value of separation factor is 0.508 and it indicates favorable biosorp-
tion (0 < RL < 1) of nickel onto sargassum tenerrimum powder.
4.3 Temkin isotherm

The Temkin and Pyzhev isotherm (King et al., 2007) equation describes the behavior of many
adsorption systems on the heterogeneous surface and it is based on the following Equation:
qe = RT ln(ATCe)/bT (6)
The linear form of the Temkin isotherm can be expressed as:
Figure 8. Langmuir isotherm for biosorption of nickel.
597

qe = (RT/ bT ) ln(AT) + (RT/bT) ln(Ce) (7)
where
R = Universal gas constant (8.314 J/mol.K)
T = Temperature of dye solution, K
AT, bT = Temkin isotherm constants
AT = exp [b(0) × b(1) / RT] (8)
b(1) = RT/ bT is the slope

b(0) = ( RT/bT ) ln (AT) is the intercept and b = RT/b (1). The Temkin isotherm is applied to
the present data and the linear plot is shown in Figure 9. The Equation obtained is:
qe = 0.606 lnCe + 1.154 (9)
The resulting bT and AT values are 4,157, A and 2.16 respectively.
4.4 Redlich-Peterson isotherm

Redlich and Peterson (1959) proposed a three-parameter isotherm to incorporate features of
both the Langmuir and Freundlich equations. It can be described as follows:
AC e
qe = (10)
1 + BC e g
where A (L/g) and B (L/ mg) are the Redlich-Peterson isotherm constants and ‘g’ is the
Redlich-Peterson isotherm exponent, which lies between 0 and 1.
The linear form of the equation is:
 C 
ln  A e − 1 = g ln(C e ) + ln B (11)
 qe 
Although a linear analysis is not possible for a three-parameter isotherm, the three iso-
therm constants – A, B and g – can be evaluated from the pseudo linear plot using a trial and
error optimization method. A general trial and error procedure is applied to determine the
Figure 9. Temkin isotherm for biosorption of nickel.
598

Figure 10. Redlich-Peterson isotherm for biosorption of nickel.
coefficient of determination (R2) for a series of values of ‘A’ for the linear regression of ln (Ce)
on ln [A(Ce/qe)–1] and to obtain the best value of ‘A’ with maximum ‘R2’. Figure 10 shows the
Redlich-Peterson plot drawn between ln [A(Ce/qe)–1] and ln Ce. For the present experimental
data, the equation obtained is:
  Ce  
ln    − 1 = 1.0718 lnCe − 1.685 (12)
  qe  
The Redlich-Peterson isotherm constants, B (L/mg) and the Redlich-Peterson isotherm

exponent (g) are 0.185 L/mg and 1.0718 respectively. The correlation coefficient of 0.96 sug-
gests that the Redlich-Peterson isotherm model is suited to describe the biosorption of
nickel.
The biosorption data is well represented by the Langmuir (R2 = 0.98), Freundlich
(R2 = 0.97), Redlich-Peterson (R2 = 0.96) and Temkin (R2 = 0.95) isotherms.
5 BIOSORPTION KINETICS
5.1 First order and second order kinetics

The pseudo first order rate equation of Lagergren is:
(dqt/dt) = K1 (qe – qt) (13)
The above equation can be presented as:
∫ (dqt /(qe – qt)) = ∫ K1 dt (14)
Applying the initial condition qt = 0 at t = 0, we get:
log (qe – qt) = log qe – (K1/2.303) t (14)
The plotting of log (qe–qt) versus ‘t’ gives a straight line for first order kinetics, facilitating
the computation of adsorption rate constant (K1). If the experimental results do not follow
the above Equation, in such cases the pseudo second order kinetic equation:
(dqt/dt) = K2 (qe – qt)2 (15)

599

is applicable, where ‘K2’ is the second order rate constant.
The other form of the above equation is:
(t/qt) = (1/ K2qe2) + (1/qe) t. (16)
In the present study, the Lagergren plot of log (qe–qt) vs. ‘t’ is shown in Figure 11. The
pseudo second order rate equation plot between (t/qt) and ‘t’ is drawn in Figure 12. The
resulting equations and constants are shown in Table 1.
Figure 11. First order kinetics for biosorption of nickel.
Figure 12. Second order kinetics for biosorption of nickel.
Table 1. Kinetic equations and rate constants.
Order Kinetic equation Rate constants R2
Pseudo first order log(qe – qt) = − 7.730t – 0.561 K1 = 17.802 min-1 0.90

Pseudo second order t/qt = 0.534 t + 1.260 K2 = 0.226 g/mg-min 0.99
600

Figure 13. Van’t Hoff plot for biosorption of nickel.
Table 2. Thermodynamic parameters of biosorption for nickel.
–(ΔG), kJ/mol
C0, ΔS, ΔH,
mg/L J/(mol–°K) J/mol 283 K 293 K 303 K 313 K 323 K
5 70.23 19.85 19.85 20.55 21.26 21.96 22.66

10 68.58 19.70 19.38 20.07 20.76 21.44 22.13
20 62.99 18.34 17.80 18.43 19.06 19.69 20.32
As the correlation coefficient for the pseudo second order kinetics is 0.999, it describes
the mechanism of nickel–sargassum tenerrimum powder interactions better than first order
kinetics (R2 = 0.93).
5.2 Thermodynamics of biosorption

The Van’t Hoff’s plot for the biosorption data are shown in Figure 13.
The more negative values reflect a more energetically favorable biosorption [15]. The
negative value of ∆G indicates the feasibility of the process and the spontaneous nature of
sorption with a high affinity for nickel biosorption onto sargassum tenerrimum powder. The
positive value of enthalpy change has confirmed that the biosorption tends to be endother-
mic in nature. The values of ∆S, ∆H and ∆G obtained in the present investigation for different
initial concentrations of nickel are shown in Table 2.
6 CONCLUSION
The equilibrium agitation time for biosorption of nickel is 120 min. The optimum dosage is
18 g/L. % biosorption is increased up to pH = 4.5. The experimental data is well represented
by the Langmuir isotherm with a higher correlation coefficient (R2 = 0.98). The biosorption
of nickel is better described by pseudo second order kinetics (K2 = 0.226 g/(mg-min)). The
biosorption is endothermic as ∆H is positive, irreversible as ∆S is positive, and spontaneous
as ∆G is negative.
601

REFERENCES
Aksu, Z. & Donmez, G. (2006). Binary biosorption of cadmium (II) and nickel (II) onto dried Chlo-
rella vulgaris: Co-Ion effect on mono-component isotherm parameters. Process Biochemistry, 41(4),
860–868.
Al-Rub, F.A., El-Naas, M.H., Benyahia, F. & Ashour, I. (2004). Biosorption of nickel on blank alginate
beads, free and immobilized algal cells. Process Biochemistry, 39(11), 1767–1773.
Bai, R.S. & Abraham, T.E. (2002). Studies on enhancement of Cr (VI) biosorption by chemically modi-
fied biomass of Rhizopus nigricans. Water Research, 36(5), 1224–1236.
Congeevaram, S., Dhanarani, S., Park, J., Dexilin, M. & Thamaraiselvi, K. (2007). Biosorption of chro-
mium and nickel by heavy metal resistant fungal and bacterial isolates. Journal of Hazardous Materi-
als, 146(1–2), 270–277.
Dàvila-Guzmàn, N., Cerino-Cordova, F., Rangel-Méndez, J. & Diaz-Flores, P. (2011). Biosorption of
lead by spent coffee ground: Kinetic and isotherm studies. In AIChE Annual Meeting, Conference
Proceedings, 1–9.
Freundlich, H. (1907). Über die Adsorption in Lösungen (Adsorption in solutions). Z. Physiol. Chem.,
57(1), 384–470.
Gupta, V.K., Rastogi, A. & Nayak, A. (2010). Biosorption of nickel onto treated alga (Oedogonium
hatei): Application of isotherm and kinetic models. Journal of Colloid and Interface Science, 342(2),
533–539.
King, P., Rakesh, N., Beenalahari, S., Kumar, Y.P. & Prasad, V.S.R.K. (2007). Removal of lead from
aqueous solution using Syzygium cumini L.: Equilibrium and kinetic studies. Journal of Hazardous
Materials, 142(1–2), 340–347.
Koedrith, P., Kim, H., Weon, J.I. & Seo, Y.R. (2013). Toxicogenomic approaches for understanding
molecular mechanisms of heavy metal mutagenicity and carcinogenicity. Int J Hyg Environ Health,
216(5), 587–598.
Krishna, R.H. & Swamy, A.V.V.S. (2011). Studies on the removal of Ni (II) from the aqueous solutions
using powder of mosambi fruit peelings as a low cost adsorbent. Chemical Sciences Journal, 2011,
1–13.
Langmuir, I. (1918). The adsorption of gases on plane surfaces of glass, mica and platinum. J. Am.
Chem. Soc., 40(9), 1361–1403.
Mata, Y.N., Blazquez, M.L., Ballester, A., Gonzalez, F. & Munoz, J.A. (2008). Characterization of the
biosorption of cadmium, lead and copper with the brown alga Fucus vesiculosus, Journal of Hazard-
ous Materials, 158(2–3), 316–323.
Nourbakhsh, M.N., Kiliçarslan, S., Ilhan, S. & Ozdag, H. (2002). Biosorption of Cr6+, Pb2+ and Cu2+
ions in industrial waste water on Bacillus sp. Chem Eng J, 85(2–3), 351–355.
Nuhoglu, Y. & Malkoc, E. (2009). Thermodynamic and kinetic studies for environmentally friendly Ni
(II) biosorption using waste pomace of olive oil factory. Bioresource Technology, 100(8), 2375–2380.
Özer, A., Gürbüz, G., Çalimli, A. & Körbahti, B.K. (2008). Investigation of nickel (II) biosorption on
Enteromorpha prolifera: Optimization using response surface analysis. Journal of Hazardous Materi-
als, 152(2), 778–788.
Pahlavanzadeh, H., Keshtkar, A.R., Safdari, J. & Abadi, Z. (2010). Biosorption of nickel (II) from aque-
ous solution by brown algae: Equilibrium, dynamic and thermodynamic studies. Journal of Hazard-
ous Materials, 175(1–3), 304–310.
Pinto, P.X., Al-Abed, S.R. & Reisman, D.J. (2011). Biosorption of heavy metals from mining influenced
water onto chitin products. Chem Eng J, 166(3), 1002–1009.
Rangsayatorn, N., Upatham, E.S., Kruatrachue, M., Pokethitiyook, P., & Lanza, G.R. (2002). Phytore-
mediation potential of Spirulina (Arthrospira) platensis: Biosorption and toxicity studies of cad-
mium. Environmental Pollution, 119(1), 45–53.
Redlich, O.J.D.L. & Peterson, D.L. (1959). A useful adsorption isotherm. J Phys Chem, 63(6), 1024.
Wang, J. & Chen, C. (2009). Biosorbents for heavy metals removal and their future. Biotechnol. Adv.,
27(2), 195–226.
602

Automation of IGCC power plant using Yokogawa DCS
K.J. Jyothir Rose & R. Anjana

Lydia Jenifer
Yokogawa India Ltd., Bangalore, India
ABSTRACT: Integrated Gasification Combined Cycle (IGCC) power plants which are
working on high-efficiency coal gasification technologies, are operated commercially or semi
commercially worldwide. Various coal gasification technologies are embodied in these plants
including different coal feed systems (dry or slurry), fireproof interiors walls (fire brick or
water-cooled tubes), oxidants (oxygen or air), and other factors. These designs which are
several decades old, but using new systems and cycles are emerging to further improve the
efficiency of the coal gasification process. The development of Distributed Control System
(DCS) for automated operation, monitoring and control functions which combines Human
Machine Interface (HMI), interlocks, logic solvers, historian, common database, report gen-
eration, alarm management and a common engineering suite into a single automated system.
The implementation is done using automatic methods of distribution which guarantee the
preservation of behavior of the whole system. The purpose of the project is to develop a suit-
able control strategy using DCS for the futuristic power plant for better monitoring, opera-
tion and availability.
1 INTRODUCTION
Integrated Gasification Combined Cycle (IGCC) is a power plant in which a gasification

process provides syngas to a combined cycle under an integrated control system. It has
potentially many advantages including high thermal efficiency, good environmental charac-
teristics, reduced water consumption etc. Hence Gasification based Power plants are future
of power production from coal or biomass. The aim of the project is to develop a Distributed
Control System (DCS), which are dedicated systems used in manufacturing processes that
are continuous or batch-oriented which is used for monitoring and controlling of an IGCC
power plant which will definitely play a significant role in future.
Studying different controlling methods for plant instrumentation and control and for this
project implementation DCS have been opted considering the advantages of DCS over other
systems. Proprietary interconnections and communications protocol is used in DCS which
uses custom designed processors as controllers. Input and output modules form component
parts of the DCS. The input modules receive information from input instruments in the proc-
ess (or field) and transmit instructions to the output instruments in the field. Computer buses
or electrical buses are used to connect the processor and modules through multiplexer or de
multiplexer. Buses also connect the distributed controllers with the central controller and
finally to the Human machine interface (HMI) or control consoles. An HMI is a software
application that presents information to an operator or user about the state of a process, and
to accept and implement the operators control instructions.
603

2 IGCC TECHNOLOGY
IGCC has potentially many advantages including high thermal efficiency, good environmen-
tal characteristics, reduced water consumption etc. Hence Gasification based Power plants
are future of power production from coal or biomass. The aim of the project is to develop a
Distributed Control System for monitoring and controlling of an IGCC power plant which
will definitely play a significant role in future. The fuel Syngas is generated using a high effi-
ciency power generation technology which gasifies coal and combustion takes place in high
efficiency gas turbine (GT) in IGCC power plant. Compared with conventional pulverised
coal (PC) fired power plants IGCC has potentially many advantages including:
High thermal efficiency: The Shell gasifier efficiency of IGCC generation is estimated
to be 46–47% net, low heating value (LHV) basis (44–45% net, high heating value (HHV)
basis), for FB class gas turbine using bituminous coal. The highest reported efficiency
for an IGCC is Efficiency of 41.8% HHV basis. Good environmental characteristics that
match or exceed the latest PC plants. The plant’s high thermal efficiency means that emis-
sions of CO2 are low per unit of generated power. In addition, emissions of SOx and
particulates are reduced by the requirement to deep clean the syngas before firing in the
gas turbine.
Reduced water consumption: IGCC uses less water. Since 60% of its power is derived
from an air based Bray ton cycle reducing the heat load on the steam turbine condenser to
only 40% of that of an equivalent rated pulverised coal fired plant. Additionally, through the
direct de-sulfurization of the gas, IGCC does not require a large flue gas de-sulfurization unit
which consumes large amounts of water, thereby reducing water consumption in comparison
with a conventional pulverised coal fired power plant. Further gains in reducing water use
can be achieved when CCS is incorporated into the plant.
Table 1. Evolution of Yokogawa DCS.
Systems Year Platform
Centum/Centum V 1975 MS-DOS based

Centum-XL/Micro-Xl 1988 MS-DOS/UNIX based
Centum CS 1993 UNIX based
CS 3000 1998 Windows based
CENTUM VP 2008 Windows based
Figure 1. Basic IGCC diagram without CO2 Capture.
604

An IGCC system is composed of a gasification unit, an air separation unit (ASU), a syn-
gas purification unit, and a combined cycle involving a gas turbine and a steam turbine.
A simplified version of a coal-fuelled IGCC cycle is shown in Figure 1.
2.1 Yokogawa CENTUM VP

A distributed control system (DCS) is a control system used in various industries as manu-
facturing system, process or any kind of dynamic system, in which the there is no central
control location for controller elements but are distributed throughout the system for one or
more controllers for each component sub-system controlled. The DCS used in the project is
YOKOGAWA DCS.
3 P&ID DESIGN AND DEVELOPMENT
This approach requires a great deal of input from the various design engineers before all of
the details have been worked out. It recommends the members of the design team to consider
all of the problems involving successful instrumentation operation. The limitation to the
detailed P&IDs design before detailed layout is complete is that the P&IDs must synchronise
with requirements in the electrical, process, instrumentation and piping as closely as possible.
If there are major modifications to be implemented to the project during detailed process
and instrumentation diagram (P&ID) development, the P&IDs must be modified. If the
P&IDs does not sync with the work of the various departments, design team members may
use incorrect information.
The second approach is to allows the P&IDs to show the instrumentation connections
only. Instrumentation designers and engineers (and possibly the electrical engineers to dou-
ble check instrumentation wiring requirements) use this approach when P&IDs are only
used among them. These diagrams do not detail the same level of information as the first
approach. The intention is to show how instrumentation and the process are related which
possibly show the electrical requirements.
4 PROCESS EXPLANATION AND METHODOLGY
One of the most popular air separation process used is cryogenic air separation, frequently
in medium to large scale plants. This technology is mostly preferred for producing nitrogen,
oxygen, and argon as gases and/or liquid products and supposed to be the most cost effective
for high production rate plants. The process of cryogenic air separation is studied and P&ID
is developed as shown in Figure 2.
Similarly the IGCC is divided into several sub sections based on operations and P&ID
for each section is developed separately. The control loops are identified and developed are
shown in Figure 3.
The Coal Grinding System provides a means to prepare the coal as a slurry feed for the
gasifier. Coal is continuously fed to the Coal Weigh Feeder, which regulates and weighs the
coal fed to the Grinding Mill. The unloading of coal, its crushing, storage and filling of
boiler bunkers in a thermal power station is covered by The coal handling plant (CHP). The
main function of coal handling unit is crushing of coal into very fine particles for gasifica-
tion and regulate the ratio of the coal and lime mix with the requirements. The lime ratio is
increased to decrease the SOx level in the syngas produced from gasification chamber.
The process of gasification to produce combustible gas also known as syngas or producer
gas from organic feeds is used. The gasification of Biomass is a thermo-chemical process that
produces relatively clean and combustible gas through pyrolytic and reforming reactions. The
product of gasification is a combustible synthesis gas, or syngas. Because gasification involves
the partial, rather than complete, oxidization of the feed, gasification processes operate in an
605

Figure 2. P&ID developed for air seperation unit.
Figure 3. P&ID developed for air separation unit.
oxygen-lean environment. The ratio of the combustible hydrogen (H2), methane (CH4), car-
bon monoxide (CO) and moisture determines the heating value of the obtained fuel.
The systems used to remove some particulates and/or gases from industrial exhaust
streams are known as Scrubber system which is a group of air pollution control devices. Gas
is contaminated and has high temperature (500–800°C) as it leaves atmospheric fluid gasifier
which focuses mostly on tar elimination and dust removal. Water or specific organic liquid
or both may be applied in tar elimination from gas. Boiling point (volatility), availability, and
price of organic liquid are major criteria for selection of a proper material. From gasifier,
606

syngas is produced and is fed into the scrubber to remove particulates and ash formed from
the gasification process.
Gas separation allows H2 to be isolated from syngas and used as a clean fuel or feedstock in
chemical production. The CO2 extracted from the process can be captured and used in chemi-
cal production or sequestered, rather than released into the atmosphere. Conventional options
for CO2 removal, such as solvent-based absorption, are energy intensive and require cooling
of syngas. The newly purified H2 must be re-pressurized and/or reheated, imposing additional
energy penalties for many applications. The exothermic shift reaction (or water gas shift reac-
tion) transfers the fuel heating value from CO to H2 and transfers the carbon from CO to CO2.
Depending on the gasification process used, the temperature of syngas leaving a gasifier can
be as high as 1600°C, recovery of heat from the high temperature syngas is essential for attain-
ing high process efficiency. Using heat recovery systems, depending on the technology used a
significant portion (5–25%) of the energy in the feed can be utilised. The syngas cooler is one of
the most crucial and highly loaded components in gasification plants. It operates with gas inlet
temperatures ranging from 1600°C to 400°C and gas-side pressures up to 8 MPa.
Removing Hg from the syngas prior to combustion is more effective in IGCC systems. This
may already occur, to some extent, via the acid gas scrubbing system, but more data are required
to verify this. The process of Syngas removal has the advantages like higher mercury concen-
tration, lower mass flow rates, and higher pressure than the stack gas. HgSIV, is a molecular
sieve (MS) that removes very low levels of elemental mercury from natural gas or syngas via a
regenerable adsorption process. It uses a 2-bed thermal-swing MS adsorption system.
Sulfur is naturally present as an impurity in fossil fuels. When the fuels are burned, the
sulfur is released as sulfur dioxide—an air pollutant responsible for respiratory problems and
acid rain. The fuel processors have to remove the sulfur from both fuels and exhaust gases as
Environmental regulations have increasingly restricted sulfur dioxide emissions. This multi-
step process has low operating costs but high capital costs too expensive for plants recovering
less than about 20 tons of sulfur per day. These plants use liquid-phase reduction-oxidation
(redox) processes to remove sulfur content from the syngas produced.
The separation and capture process for producing CO2 stream comprehensively includes
all the operations that take place at the power plant site, including compression. For ease of
transport, CO2 is generally compressed to the order of 100 atm. The NOx and SO2 should be
cleaned up prior to CO2 separation as required by flue gas approach in use today. Integrated
coal gasification combined cycle (IGCC) plants are an example of the hydrogen route.
The Heat Recovery Steam Generation (HRSG) receives the exhaust gases from the GT
discharge. The exhaust gas, flowing in counter flow with respect to the steam/water coils,
cools down by transferring heat to steam/water. The flue gas temperature at the stack is about
230°F (110°C), even though lower temperatures [200°F (93°C)] can be used if the fuel gas is
very clean and Sulfur-free. The HRSG is, therefore, similar to a heat exchanger in which the
shell side carries the flue gas and the various sections of the tube side carry steam or water. It
has also the characteristics of a boiler because there are one or more steam drums, where the
generated steam is separated from boiling water before entering the super heaters.
5 RESEARCH FINDINGS AND EXPLORATION
The objective of this paper is to develop a DCS for IGCC Power Plant. Initially the com-
plete processes of Integrated Gasification Combined Cycle (IGCC) Power plant is studied and
found out the requirements for controlling the processes and the hardware used for implement-
ing the control system for different parameters like pressure, temperature, flow, level etc. The
whole power plant consists of several sections of which each section is studied and P&ID for
the process is developed. The P&ID is a specialized document shown on a side view represen-
tation of all equipment. Integration and properly designed interfacing between the DCS and
other digital control packages is essential. The serial links should be made redundant to ensure
the maximum operating continuity. The system bus and the input/output (I/O) buses are also
implemented to be redundant, for guaranteeing the maximum uptime. The sequence of events
607

control function carried out in the DCS, the appropriate I/O cards have to be correctly antici-
pated or a dedicated system must be connected to the DCS via a serial link. The next step is to
identify the control loops and function blocks required to realise the control strategies required
for the project. After this two procedures the whole plant has to be displayed and controlled
using Human Machine Interface (HMI). Next Step is to develop the Graphics section to repre-
sent the whole plant and to view different operations and to control each controllers. In graph-
ics whole process of the plant can be visualized. The Alarm Processing is a message processing
function performed by the HMI when alarm occurs during plant operation. The issued alarm
can be used by the operator to determine what action is to be taken for the abnormality and
then process the alarm. The operator can perform an acknowledgment action for the alarm
output. These alarms are identified and implemented in the project.
The graphical user interface to interact with the functioning and overall processes of the
plant. From the graphical interface the user can know the present position or value of the
controller and the user can take necessary actions to change the values such as set value or
put a motor to on/off position depending on the situations arising. The user can monitor the
Figure 4. Graphics developed for the ASU of power plant.
Figure 5. Controller action with varying PV.
608

functions of the plant from the interface. The representations in the interface can be con-
figured during the designing of the graphics development. The graphics developed for each
section is shown in Figure 4.
The controller action for different values of Process Variable (PV) using field instruments
is shown in Figure 5.
6 CONCLUSION
The proposed control system using Yokogawa CENTUM VP and procedure presents an effec-
tive approach for monitoring and control. In real time the implementation of this method is
very effective and system is available almost 99% time. The design of P&ID of the system
developed by carefully studying the operations and plant processes of IGCC Power Plant.
The present study has demonstrated the design of DCS by I/O station inputs from the
different equipments. The quantities are measured and controlled, and control valves of the
processes are manipulated in real time to implement temperature, pressure, level and flow
rate control, breakdowns are detected and the system is maintained. Live measured values
and status indications reveal the current situation. Process operators monitor and control the
long-distance processes from the console.
ACKNOWLEDGMENT
This work is mainly carried out at Yokogawa India Ltd. Bangalore, India. I would like to
thank all co-authors for the important discussions about the work.
REFERENCES
[1] Sher shah Amarkhail, ‘Air Separation’, Development of human resource capacity of Kabul poly-
technic university”.
[2] Monika Kurková-Zdeněk Klika-Petr Martinec-Jaroslava Pěgřimočová, ‘Composition of bitumi-
nous coal in dependence on environment and temperature of alteration’, Kurkova, M. et al. 2003.
[3] Anil Bose, ‘Classification of Coal in India’, Indian Geography.
[4] Jeffery Phillips, ‘Different types of gasification and their integration with gas turbines’, EPRI/
Advanced Coal Generation.
[5] Younes Chhiti, Mohammed Kemiha, ‘Thermal Conversion of Biomass, Pyrolysis and Gasifica-
tion’, The IJES.
[6] Yongseung Yun, Seung Jong Lee and Seok Woo Chung, ‘Considerations for the Design and Opera-
tion of Pilot-Scale Coal Gasifiers’, Institute for Advanced Engineering, Suwon, Republic of Korea.
[7] Marek Balas, Martin Lisy, Zdenek Skala, Jiri Pospisil, ‘Wet scrubber for cleaning of syngas from
biomass gasification’, Indiana Council of Administrators of Special Education.
[8] Andrej Lotrič, ‒ Mihael Sekavčnik ‒ Christian Kunze ‒ Hartmut Spliethoff, ‘Simulation of Water-
Gas Shift Membrane Reactor for Integrated Gasification Combined Cycle Plant with CO2 Cap-
ture’, Strojniški vestnik – Journal of Mechanical Engineering.
[9] J.E. Jamison, B.G. Lipták, A. Rohr, ‘Power Plant Controls: Cogeneration and Combined Cycle’.
[10] Hyungwoong Ahn, Zoe Kapetaki, Pietro Brandani, Stefano Brandani, ‘Process simulation of a
dual-stage Selexol unit for pre-combustion carbon capture at an IGCC power plant’.
[11] Howard Herzog, ‘An Introduction to CO2 Separation and Capture Technologies’, MIT Energy
Laboratory.
[12] Alan Darby, “Hydrocarbon upgrading gasification program’, Alberta Energy Research Institute
Grant.
[13] Steve Fusselman, Alan Darby and Fred Widman, “Advanced gasifier pilot plant concept defini-
tion”, September 2006.
[14] Yongseung Yun, Seung Jong Lee and Seok Woo Chung, ‘Considerations for the Design and Opera-
tion of Pilot-Scale Coal Gasifiers’.
609

[15] Kosan Roh and Jay H. Lee, ‘Selection of Control Structure of Elevated Pressure Air Separation
Unit in an IGCC’.
[16] Makarand M. Joshi, ‘Development of Condition Based Maintenance for Coal Handling Plant of
Thermal Power Stations’, Plant Maintenance Resource Center.
[17] Taylor and Francis Group, ‘Combined Cycle Power Plants’.
[18] Everret B Woodruff, Herberrt B Lammers, Thomas F Lammers, ‘Steam Plant Operations’.
[19] Van Dijka, K. Damenb, M. Makkeec, C. Trappd, ‘Water-gas shift (WGS) operation of pre-
combustion CO2 capture pilot plant at the Buggenum IGCC’, H.A.J.
[20] Qian Zhu, ‘High Temperature Syngas coolers’, IEA Clean Coal Centre.
[21] John Markovs, ‘Optimized Mercury Removal In Gas Plants’, Adsorption Solutions LLC Cross
Junction, Virginia, U.S.
610

Deprotonation studies on polyaniline polymethyl methacrylate

blends processed from formic acid
V.O. Rejini, A.H. Divya & Rakesh S. Nair

ABSTRACT: Polyaniline (PANI) and Polymethylmethacrylate (PMMA) blend films were

prepared from homogeneous solutions in formic acid, which serves both as a dopant as well
as a solvent. The absorption spectra revealed that formic acid processed PANI and its blends
with PMMA to be deprotonated in atmospheric conditions. Stability studies were done in
three ways: with and without an additional protonating agent, Camphoursulphonic Acid
(CSA), and with an addition of a plasticizer to the host matrix. The results showed that the
presence of CSA imparts stability or resistance against deprotonation.
Keywords: Polyaniline; PMMA; solution blending; blends; plasticizer; deprotonation
1 INTRODUCTION
Solution processing has been reported to be a quite efficient method when both polyaniline
and the polymer matrix are ‘compatible’ with each other and ‘soluble’ in a common solvent
(Barra et al., 2002). The conductivity of solution-cast Polyaniline (PANI)/polymer blends
depends upon the ability of the solvent to finely disperse the conducting polymer, and on
the flocculation of the dispersed PANI in the blend (Paul & Pillai, 2002). The deprotona-
tion nature or the stability of the dopant also play a vital role on the overall conductivity.
Most polyaniline blends described in the literature were processed from m-cresol, a solvent
which is both acidic and high boiling point. If the emeraldine base and the host polymer are
co-soluble in an acid dopant, the blend can be obtained as a film in the conducting form by
solution casting.
The conductive PANI/polyamide-11 blend fibers were prepared by wet-spinning technol-
ogy from concentrated sulfuric acid with relatively high electrical conductivity (Zhang et al.,
2001). High strength and high modulus electrically conducting PANI composite fibers were
also reported (Hsu et al., 1999) from PANI/PPD-T (poly (para-phenylenediamine) tereph-
thalic acid) sulfuric acid solutions. Due to the ease of handling and solvent removal, it is
more convenient to use liquid organic acids than sulfuric acid as solvents. Abraham et al.,
(1996) used formic acid as the solvent as well as dopant for a polyaniline–nylon 6 blend sys-
tem. The chemical modification or blending of polyaniline with nylon 6 does not affect the
crystal structure of either polyaniline or nylon 6. This was confirmed by X-ray diffraction.
The maximum conductivity of the films was about 0.2S/cm, corresponding to a weight ratio
of 0.5 (w/w) for PANI and Nylon 6. Formic acid was also used by Anand et al. (2000) for the
preparation of blends of PANI derivatives (Poly (O-Toluidine) (POT), Poly (M-Toluidine)
(PMT)) with Polymethylmethacrylate (PMMA). The blend was precipitated by the addition
of the formic acid solution to water (non-solvent). The thermal stability of the blends was
reported to be greater than that of their respective salts. Zagórska et al. (1999a) studied the
stability against deprotonation of polyaniline/polyamide 6 blends processed from formic acid.
They prepared polyaniline-polyamide 6 blends in two different ways: one with an additional
protonating agent and the other without an additional protonating agent. The blends of
polyaniline and polyamide 6 processed from formic acid were prepared without an additional
611

protonating agent. Performed stability studies found them to be unstable at ambient labora-
tory conditions and to have a tendency to deprotonate and gradually lose their conductivity,
which is in contradiction to the observation made by Abraham et al. (1996). UV-VIS spectra
of polyaniline/polyamide 6 blend films were collected for increasing exposure time to labora-
tory atmosphere. They verified the deprotonation behavior by instantaneous increase of the
peak 630 nm, which is the characteristic of non-protonated PANI. An addition of a support-
ing protonating agent to the polyaniline/polyamide 6 systems improved their environmental
stability with respect to the deprotonation process. Anand et al. (2000) also reported UV-
visible spectra of poly (o- and m-toluidine) polymethylmethacrylate blends, which showed
only two peaks at 315 and 610 nm for all blend compositions; this is the characteristic of a
nonprotonated state. The absorption spectrum of formic acid processed POT-PMMA blends
in Dimethyl Sulphoxide (DMSO) exhibited two bands around 315 and 610 nm.The indica-
tion of insolubility of POT salt in DMSO and the presence of the base in the salt solution
were indicated. There was no mention of the environmental instability of formic acid proc-
essed systems. Juvin et al. (1999) studied conductive blends of polyaniline with plasticized
polymethylmethacrylate and reported that plasticizer dibutyl phthalate increases interaction
between the blend components and thereby improves the conductivity of the blend. Also,
these blends show good resistance to deprotonation against basic media for all the dopants
such as Camphorsulphonic Acid (CSA) and three different polyalkylene phosphates stud-
ied. Solution-cast blends of PANI/PMMA were discussed in the literature for their electrical
properties with respect to their interaction and compatibility (Yang et al., 1993).
In the present study we used formic acid as a co-solvent for emeraldine base and PMMA.
For deprotonation studies, three types of PANI/PMMA blend films were prepared: (1) with
additional protonating agent; (2) without an additional protonating agent; and (3) with plas-
ticized PMMA. Four supporting protonating agents were reported in the literature: phenyl-
phosphonic acid, camphorsulphonic acid, 2-acrylamide-2-methyl-1-propane sulphonic acid
and dibutyl phosphate. This do not promote phase separation. We used camphorsulphonic
acid as an additional protonating agent. Formic acid is known to be a very good solvent for
CSA doped polyaniline (Cao et al., 1992). The additional/supporting protonating agent would
exchange with formic acid as a polyaniline dopant at the final stage of the composite casting.
2 EXPERIMENTAL
2.1 Materials
The monomer aniline was double-distilled under reduced pressure prior to use. Ammonium
persulphate, ammonium hydroxide, formic acid and hydrochloric acid are analytical grade
reagents and used without purification. PMMA was supplied by SUMIPEX, Korea.
2.2 Synthesis of polyaniline

PANI powder in emeraldine form was synthesized by chemical oxidation of aniline with
ammonium persulphate ((NH4)2S2O8) as the oxidant in 1M aqueous hydrochloric acid solu-
tion (Angelopoulos et al., 1988). The mixture was constantly stirred in an ice bath for 6 h.
The precipitate collected was washed, first with aqueous hydrochloric acid and then with
acetone. PANI salt was dedoped using 0.2 M NH4OH.
2.3 Preparation of PANI/PMMA blends

The approximate solubility of emeraldine base in formic acid was evaluated using the pro-
cedure reported in the literature (Zagórska et al., 1999a). The emeraldine base and PMMA
(10% wt/vol.) were dissolved separately in formic acid. The blend mixtures were prepared
by varying the weight ratios of emeraldine base to PMMA. A specific composition of 10/90
wt% of polyaniline and PMMA respectively were prepared. The mixtures were sonicated for
612

ten min. and then magnetically stirred for 6h at room temperature. The resulting homogene-
ous solution was made into films by casting. The blend films obtained were dried for 48 h
under vacuum. The blends are denoted by using the initial weight percentage of emeraldine
base and PMMA.
2.4 Preparation of PANI/PMMA blends with an additional protonating agent

A specific composition of polyaniline–polymethylmethacrylate (10/90) blends were prepared
by co-solvation method with an additional protonating agent. CSA as an additional proto-
nating agent was added in the molar ratio of 0.5 per PANI repeat unit, involving one ring and
one nitrogen atom (Zagórska et al., 1999b). Blend films were prepared by solution casting.
2.5 Preparation of PANI with plasticized PMMA blends

Polymethyl methacrylate (10% wt/vol.) in formic acid was plasticized using dimethyl phtha-
late (35 wt%). Plasticized PMMA solution was mixed with polyaniline solution in formic acid
for 10/90 wt% (PANI/PMMA) composition and films were prepared by solution casting.
2.6 Deprotonation studies

For deprotonation studies, blend films of a particular composition (10/90) were exposed to
laboratory atmosphere (29°C) for seven days. UV spectra were measured every day at a par-
ticular time for seven days. The effect of additional protonating agent, CSA, on PANI/PMMA
blend system was determined and also any influence was found on using plasticized (dimethyl
phthalate) PMMA on the deprotonation behavior of the formic acid doped polyaniline system.
3 CHARACTERIZATION
The UV-VIS absorption spectra of the blend films were recorded using a Varian Cary 5E
model UV-VIS near–IR spectrophotometer.
The base form shows two major peaks at 630 and 330 nm, which is the characteristic absorp-
tion spectrum of the base form of the PANI. The peak at 330 nm regions is assigned to the
Figure 1. UV-VIS spectra of the salt form and base form of polyaniline.
613

Figure 2. UV-VIS spectra collected for PANI/PMMA blend film, exposed to the laboratory atmos-
phere for seven days.
Figure 3. UV-VIS spectra of PANI/PMMA blends with camphoursulphonic acid.
Figure 4. UV-VIS spectra of formic acid doped polyaniline with plasticized polymethylmethacrylate
blend film, exposed to atmosphere for seven days.
614

Π-Π* transition, related to the extent of conjugation between adjacent phenyl rings in the
polymer chain (Zhang et al., 2001). The peak at 630 nm is a measure of extended conjugation,
corresponding to the excitation-like transition from the highest occupied benzenoid ring to
the lowest unoccupied quinoid ring, caused by interchain or intrachain charge transfer. For
fully protonated PANI upon UV irradiation, the intensity of the 320 nm band decreases and
a new absorption band appears at 440 nm, assigned to the polaron transition. An absorption
tail extending into the near-IR region is also observed. These features are characteristic of
protonated PANI. The absorption value of peaks at 950 nm increased with an increase of
protonation state. The peak at 630 nm indicates that the quinoid segment on the polyemer-
aldine chain is present in partially protonated state, or that formic acid doped polyaniline
undergoes slow deprotonation.
The absorption spectra of the blend film indicate that formic acid polyaniline undergoes
deprotonation as time proceeds. Within seven days, the blend film was completely deproto-
nated, the peaks at 440 nm and 900 nm disappear and a specific peak at around 600 nm appears.
In the presence of an additional protonating agent, CSA, the system retains its conductiv-
ity or stability against deprotonation, even after being exposed to the laboratory atmosphere
for seven days. Absorbance spectra show energetic peaks at 430 nm (related to the ∏–∏*
transition on the polymer chain) and around 800 nm, attributed to polaron transitions, indi-
cating the presence of polyaniline in the salt form.
Absorbance spectra indicate that the presence of the plasticizer does not make any
change to the deprotonating nature of the formic acid processed system. The blend film was
deprotonated within seven days. The PANI-CSA-plasticized PMMA system was reported
for its improved conductivity and resistance against deprotonation rate (Barra et al., 2002).
5 CONCLUSIONS
Deprotonation studies show that PANI/PMMA blends and PANI-plasticized PMMA blends
processed from formic acid are not environmentally stable. The interaction between the blend
components and compatibility do not impart any resistance to deprotonation. As per the
literature, formic acid doped PANI with nylon 6 blends are immiscible and PANI/PMMA
blend systems are highly compatible. From the results obtained, we could conclude that the
nature of host matrix and miscibility or immiscibility of the blend components do not make
any change in the deprotonating nature of formic acid processed polyaniline blend systems.
PANI/PMMA formic acid system with an additional protonating agent, camphoursulphonic
acid, showed good stability against deprotonation.
REFERENCES
Abraham, D., Bharathi, A. & Subramanyam, S.V. (1996). Highly conducting polymer blend films of
polyaniline and nylon 6 by co-solvation in an organic acid. Polymer, 37(23), 5295–5299.
Anand, J., Palaniappan, S. & Sathyanarayana, D.N. (2000). Solution blending of poly (o- and m-
toluidine) with PMMA in formic acid medium: Spectroscopic, thermal and electrical behaviour. Eur
Polym J, 36(1), 157–163.
Angelopoulos, M., Asturias, G.E., Ermer, S.P., Rey, A., Scherr, E.M., Macdiarmid, G., & Epstein, A.J.
(1988). Polyaniline: Solutions, films and oxidation state. Molecular Crystals and Liquid Crystals,
160(1), 151–163.
Barra, G.M., Levya, M.E., Soares, B.G. & Sens, M. (2002). Solution-cast blends of polyaniline-DBSA
with EVA copolymers. Synth Met, 130(3), 239–245.
Cao, Y., Smith, P. & Heeger, A.J. (1992). Counter-ion induced processibility of conducting polyaniline
and of conducting polyblends of polyaniline in bulk polymers. Synth Met, 48(1), 91–97.
Cao, Y. & Smith, P. (1993). Liquid-Crystalline solutions of electrically conducting polyaniline. Polymer,
34(15), 3139–3143.
Dan, A. & Sengupta, P.K. (2004). Synthesis and characterization of polyaniline prepared in formic acid
medium. J Appl Polym Sci, 91(2), 991–999.
615

Hsu, C.H., Shih, H., Subramoney, S. & Epstein, A.J. (1999). High tenacity, high modulus conducting
polyaniline composite fibers. Synth Met, 101(1–3), 677–680.
Juvin, P., Hasik, M., Fraysse, J., Planès, J., Pron, A. & Kulszewicz-Bajer, I. (1999). Conductive blends of
polyaniline with plasticized poly (methyl methacrylate). Appl Polym Sci, 74(3), 471–479.
Paul, R.K. & Pillai, C.K.S. (2002). Melt/solution processable conducting polyaniline: Elastomeric
blends with EVA. J Appl Polym Sci, 84(7), 1438–1447.
Terlemezyan, L., Mihailov, M. & Ivanova, B. (1992). Electrically conductive polymer blends comprising
polyaniline. Polymer Bulletin, 29(3–4), 283–287.
Yang, C.Y., Cao, Y., Smith, P. & Heeger, A.J. (1993). Morphology of conductive, solution-processed
blends of polyaniline and poly (methyl methacrylate). Synth Met, 53(3), 293–301.
Zagórska, M., Harmasz, E., Kulsewicz-Bajer, I. Proń, A. & Niziol, J. (1999a). Blends of polyaniline with
polyamide 6 processed from formic acid. Synth Met, 102(1–3), 1240.
Zagórska, M., Taler. E, Kulsewicz-Bajer, I., Proń, A. & Niziol, J. (1999b). Conductive polyaniline-
polyamide 6 blends processed from formic acid with improved stability against deprotonation. J Appl
Polym Sci, 73(8), 1423–1426.
Zhang, Q., Jin, H., Wang, X. & Jing, X. (2001). Morphology of conductive blend fibers of polyaniline
and polyamide-11. Synth Met, 123(3), 481–485.
616

Characterization of brushless DC motor for control valve

actuation in rocket propulsion systems
K.N. Ajeesh & Suhana Salim

R. Sujith Kumar
VSSC, Trivandrum, Kerala, India
ABSTRACT: Control valves are usually actuated by pneumatic signals. Electrical actuators
can also be used, which are more accurate and cheap. Laboratory research on a flow control
valve actuated by brushless DC (BLDC) motor has been carried out. The characterization of
a control valve driven by a DC motor has been obtained to determine whether the outlet flow
depends on motor parameters such as the motor speed, stepping angle etc. Laboratory test
for the characterization work is done by using nitrogen as the test fluid. LabVIEW is used for
the data acquisition. Different tests were carried out with varying speeds and from the time
constants obtained, first order systems were designed. The theoretical response curve for first
order systems were generated using MATLAB software and compared with responses of
controllers with conventional pneumatic actuator. The comparison showed electrical actua-
tors were much faster than pneumatic actuators.
1 INTRODUCTION
Control valves are essential components of any piping system. When we use the term valve,
it is manually operated whereas a control valve is one with an actuator that automatically
opens or closes the valve fully or partially to a position dictated by signals transmitted from
the controlling instruments. Based on actuation, control valves can be mainly classified into
quarter turn valves, multi-turn valves and check valves. Quarter turn valve allows only 90o
rotation and it includes ball valves, butterfly valves spherical valves and plug valves. In this
work a ball valve is used for characterization study. A multi-turn valve allows 360o rotation
and it requires 4–5 turns to completely open or close the valve. Gate valves, globe valves and
pinch valves are all multi-turn valves.
Apart from process industries, control valves find lot of other applications like power sta-
tions, rockets & space crafts, automobile systems etc. For such applications, sometimes it is
required to develop non-conventional type of technologies. Control valves are actuated by
using pneumatic, hydraulic and electrical actuators. Most of the industries are using conven-
tional pneumatic signals for valve actuation. Control valve can also be actuated by using DC
motor.
The objective of this work was to find the response time for a ball valve coupled with a
BLDC motor at various speeds so as to characterize the system. Knowledge about BLDC
motor and its drive is essential for the characterization study. The design and implementation
of BLDC motor drive for automotive applications was reported to give reliable results (Park
et al. 2012). Most of the rocket systems use light weight propulsion systems. One of the most
desired technological requirements for an efficient aerospace launch vehicle is the use of light
weight propulsion system. Brushless DC motors whose efficiency can be greater than 90%
are useful for this purpose in this sense (Van Neikerk 2015).
617

2 TEST SETUP
2.1 System description

A rectangular aluminium sheet is taken and it is bent into an L shaped structure. The power
supply circuit is mounted on its base. The control circuit (driver) of the motor is also mounted
near to it. Five switches and a 20 K POT are placed on the uppermost part. So for developing
the system, the hardware components required are a power supply circuit, BLDC motor and
its control unit, ball valve and two pressure transmitters. The motor requires 24V DC supply
for its operation. Control circuit for opening and closing of the valve and the BLDC motor
is shown in the Figure 1.
The valve and the actuator portion are placed on field and the DAQ and PC is placed on
the control room. Figure 2 shows the arrangement of switches along with motor-ball valve
interconnection.
The placement of upstream and downstream pressure transmitters along with valve-motor
interconnection is shown in Figure 3. This developed module was placed on the test bay. It
only possesses the valve and actuator portion.
2.2 Instrumentation scheme

The instrumentation scheme provides for measurement of pressure and flow. The details of
the sensors, sensor gauges and sensor location are given below.
a. Two strain gauge type pressure sensors with range 0–10 kg/cm2 for upstream and down-
stream pressure measurement.
b. 24V DC power supply unit for motor triggering.
Figure 1. BLDC motor and its control unit.
Figure 2. System configuration.
618

Figure 3. Valve-Motor interconnection along with pressure transmitters.
c. Signal conditioning module.

d. USB Data Acquisition module.
e. 10 volts power supply for sensor excitation.
f. Digital panel meters for continuous monitoring of the pressure lines.
The data acquisition is carried out by using a low cost USB data acquisition module and
a laptop with USB interface.
A LabVIEW based software and real time display panel was developed for parametric
display and acquisition. The software was also provided with a provision for offline analysis
of acquired data.
3 EXPERIMENT
Stored nitrogen gas at a pressure of 150 bar is regulated to 10 bar by using a spring loaded
pressure regulator. It is then passed through a solenoid valve. A pressure transducer is pro-
vided next to the solenoid valve in the flow line to measure the incoming pressure. The gas
then passes through the motorized ball valve. The upstream pressure can be measured by
using a strain gauge type pressure transducer. By using pressure transmitter, the measured
pressure is transmitted to the DAQ system. For DAQ purpose, NI data acquisition unit and
LABVIEW software is used. The details of the methodology adopted for the test and test
sequence are explained below.
3.1 Test methodology

The methodology adopted for the test includes the following steps:
a. Turn on the DAQ system.
b. Turn on the regulator. Set the regulator pressure as 10 bar.
c. Turn ON the power supply of the motor.
d. Switch on the START button.
e. Switch ON the RUN button.
f. After specified time delay switch OFF the RUN button.
g. Press STOP.
h. Turn OFF power supply of motor.-driver.
i. Turn off regulator.
j. Turn off the DAQ system.
619

Table 1. Test sequence.
Time (s) Events
T-15 Start count down

T-10 DAQ ON
T-6 GN2 supply ON
T-3 Switch on the power supply of motor
T-2 Switch on START/STOP
0 Switch on RUN/BRAKE
T+3 Switch off START/STOP
T+4 Switch off RUN/BRAKE
T+5 Switch off the power supply of motor
T+6 GN2 supply off
T+10 DAQ off
Figure 4. Block diagram of measurement system.
3.2 Test sequence

The sequence of operation adopted for the test is shown in Table 1. Six tests with different
speeds were carried out. All the tests were conducted in the test bay equipped with nitrogen
cylinders, pressure regulators, motorized valves, pressure transducers etc.
Figure 4 shows the block diagram of the measurement scheme. 10V DC Power supply is
used for sensor excitation. The pressure measurement is done by using pressure sensors.
4 RESULTS
The obtained experiment results were analysed and graphs were plotted. Time constants
obtained and first order systems were designed. Figure 5 shows first order system with dif-
ferent time constants.
4.1 MATLAB responses

For obtaining the transient characteristics such as rise time, peak time and settling time of
the response curves, MATLAB software is used. The MATLAB responses obtained for one
trial test (time constant = 0.0215 seconds) is shown in Figure 6.
The time constant, rise time, peak time and settling time obtained for six different trial
tests are given in Table 2.
Comparative study
A comparison of the results obtained in this work with a previous work using pneumatically
actuated controller (Sanoj et al. 2013) is tabulated as shown in Table 3.
620

Figure 5. First order system with different time constants.
Figure 6. MATLAB response for trial 1.
Table 2. Transient characteristics of different test trials.
Time constant (s) Rise time (s) Settling time (s) Peak time (s)
Trial 1 0.0215 0.0473 0.0841 >0.14

Trial 2 0.0266 0.0585 0.1040 >0.16
Trial 3 0.0272 0.0598 0.1060 >0.16
Trial 4 0.0199 0.0437 0.0779 >0.12
Trial 5 0.0201 0.0442 0.0786 >0.12
Trial 6 0.0303 0.0666 0.1190 >0.18
Table 3. Comparison of pneumatic and electrical actuators.
Rise time (s) Peak time (s) Settling time (s)
Electrical actuator Trial 4 0.044 >0.12 0.078

Trial 2 0.059 >0.16 0.104
Trial 6 0.066 >0.18 0.119
Pneumatic actuator* Ziegler-Nichols 0.564 1.68 5.346
Tyreus-Leyben 0.595 – 17.69
Fuzzy 1.124 2.64 1.889
Fuzzy PID 0.52 2.11 1.691
*(Sanoj et al. 2013).
621

For comparison, three trials were taken for the study. Considering the rise time, the
response characteristics of electrical actuators are about ten times faster than that of pneu-
matic actuators.
While, in the case of pneumatic actuators, even with the use of Fuzzy PID controller, the
peak time obtained was 2.10 seconds whereas in case of electrical actuator, the peak time is in
the range of 0.12 to 0.18 seconds which indicates a much better response.
Thus it is clear that valve actuation using brushless DC servo motors is much faster and
advantageous than the conventional pneumatic actuators.
5 CONCLUSION
The characterization of a ball type control valve actuated by brushless DC motor has been
carried out. Different tests were carried out with varying speeds by using an external poten-
tiometer and from the time constants obtained, first order systems were designed. The theo-
retical response curve for first order systems were generated using MATLAB software and
response characteristics such as rise time, peak time, settling time etc were found out from
this output response curve.
A comparative study performed between the motor driven valve and that of a pneumatic
valve has indicated that the response time are much lower for the motorised valve making
the valve responses much faster. Hence a motorised valve can be used for control application
that warrants faster response. Rocket engines used in propulsion systems require much faster
response, so motorized valves are found to be a very good choice for control valve actuation
in piping systems such as those used in the aerospace industries.
REFERENCES
Avila, A., Carvajal, C. & Carlos Cotrino, C. 2014. Characterization of a Butterfly-type Valve, IEEE
transactions, 16(1): 213–218.
Fisher Controls International. 2005. Control Valve Handbook.
Ireneusz, D. & Stanislaw, F. 2014. Characteristics of flow control valve with MSMA actuator, Interna-
tional Carpethian Control Conference, Krakow, Poland.
Li, T. Huang, J. Bai, Y. Quan, L. & Wang, S. 2015. Characteristics of a Piloted Digital Flow Valve Based
on Flow Amplifier, International Conference on Fluid Power and Mechatronics.
Miller, R.W. 1996. Flow Measurement Engineering Handbook. New York: McGraw Hill.
Oriental Motors, Brushless DC Motor and Driver Package, BLH Series Operating Manual.
Park, J.S. Bon-GwanGu, Kim, Jin-Hong, Choi, Jun-Hyuk & Jung, In-Soung. 2012. Development of
BLDC Motor Drive for Automotive Applications, Electrical Systems for Aircraft, Railway and Ship
Propulsion (ESARS).
Sanoj, K.P., Ajeesh, K.N., Ganesh, P. & Sujithkumar, R. 2013. Characterization of Mass Flow Control
System for Liquid Rocket Engine Application, ‘Proceedings of National Conference on Advanced
Trends in Chemical Engineering’, Govt. Engineering College, Thrissur, India.
Van Neikerk, D. 2015. Brushless Direct Current Motor Efficiency Characterization, Electrical Machines &
Power Electronics (ACEMP).
622

Sequestration of hydrogen sulfide and carbon dioxide from biogas
Antony P. Pallan & S. Antony Raja

Department of Mechanical Engineering, Karunya University, Tamil Nadu, India
B. Sajeenabeevi
Department of Chemical Engineering, Government Engineering College, Kerala, India
C.G. Varma
Department of Live Stock Production Management, Kerala Veterinary and Animal Sciences University,
Kerala, India
ABSTRACT: Anaerobic Digestion (AD) of wastes is one of the best treatment methods
in the arena of waste management. Biogas, the end product of AD, comprises of 40%–75%
CH4 and 25%–55% CO2, with other minor components such as H2S and SO2. Increased
concentration of minor gases will cause the corrosion of the pipe lines which are being used
in bio-energy generation. Hence, the present study is undertaken to develop a low-cost scrub-
bing mechanism for toxic gas removal. A biogas purification system with multi scrubbing
apparatus (Phase I and Phase II) was designed and utilized for the study. The principle of
chemical absorption was employed and the efficiency of the different caustic solutions at
saturated concentration was investigated. Scrubbing at Phase I reported a 5% increase in
methane and a 5.8% removal of CO2. The removal rate of 43% and 37% was observed for H2S
and NH3 respectively. Carbon dioxide was removed at a rate of 34.9% for KOH, followed by
NaOH and Ca(OH)2 at a rate of 34.1% and 33.9% respectively, in a time duration of three
minutes. It has been found that the absorption capacity of caustic solution was dropping
‘within’ a short time period. Hence, it is necessary to replace the caustic solution in order to
uphold the chemical at saturation point.
Keywords: anaerobic digestion, scrubbing, chemical absorption
1 INTRODUCTION
Anaerobic Digestion (AD), popularly known as biogas technology, has gained a lot of
momentum in the arena of waste management because of its ‘dual role’, that is, the conver-
sion of waste to energy and the mitigation of Green House Gas emission from the disposal
of waste. The end product of AD is mainly biogas, which is principally comprised of 40%–
75% CH4 and 25%–55% CO2 with other minor components such as H2S and SO2 [Kadam &
Panwar, 2017]. In India, biogas technology is chiefly employed in the management of manure
and farm waste, produced by the activities of agriculture and its allied sectors. These wastes
are rich in carbon and nitrogen, which is a prerequisite for the bacteria involved in AD. The
C:N ratio of 25–30 is ideal for AD [Sanaei-Moghadam et al., 2014], but due to the practices
adopted there is an alteration in the C:N ratio to be either too low or too high. This in turn
will have an effect on the metabolism of anaerobic bacteria, finally affecting the composition
of the biogas with an increased concentration of undesirable gases such as H2S, CO2 and
NH3 [Scano et al. 2014].
An increased concentration of these gases is not suitable for the combustion sys-
tems. Carbon dioxide will decrease the calorific value of the biogas because of its non-
combustible nature. It is non-toxic, but other gases present such as H2S are toxic when
623

inhaled [Tippayawong & Thanompongchart, 2010]. Thus, the condition of the biogas plant
will go down, affecting the economic returns of the farm and the health of the farmer.
Earlier, studies have also revealed that the direct utilization of the pure biogas had decreased
the efficiency of the machinery employed [Surata et al, 2014]. Hence, due to this current
situation many experimental methods were developed for the purification of biogas, which
is broadly known as a scrubbing mechanism. Each method works on a unique principle
that targets the removal of gases [Kapdi et al, 2005]. There were many models of scrubbers
developed by various researchers, but this present study was conducted to develop a holis-
tic scrubbing mechanism for harvesting purified biogas. The purified biogas, being used
for cooking purposes, electricity generation and vehicle fuel, helps to improve the world’s
energy security. The purified biogas provides the added advantage of an increased efficiency.
2.1 Theory
Chemical absorption is an adequate technology for the removal of CO2, NH3 and H2S from
the biogas. In a scrubber system, effluvium from biogas is transferred from gas to liquid as a
part of the reaction [Privalova et al., 2013]. Amines, caustic solvent and amino acid salt solu-
tions are the various chemicals which are used in biogas-purification [Abdeen et al., 2016], but
caustic solvent is mostly chosen for the purpose of cost efficiency. All aqueous solution for
chemical scrubbing was prepared at St Thomas College chemical lab, Thrissur, Kerala. The
dissolved caustic salt reacts with CO2 as part of the purification process [Üresin et al., 2015].

A 0.75 m3 capacity bio-digester (floating drum type) was utilized for this study, located at Eco
Farm, Kerala Veterinary and Animal Sciences University (KVASU), Mannuthy, Kerala. Outlet
from the digester was connected to a total biogas purification system with multi scrubbing
apparatus (Phase I and Phase II), as shown in Figure 1. Both the phases were fabricated with
PVC material, 100 cm in height and 20 cm in diameter, with an overall volume ratio of 0.031 m3.
Phase II contains a cup structured reducer (Figure 2) for efficient purification, to delay biogas
from bubbling inside the apparatus. The Phase I scrubbing unit is filled with a mixture of lime-
stone with metallic waste, to remove the majority of the NH3 and H2S. CO2 was also filtered
out to a certain extent in Phase I, reducing the cost of chemical absorbent in Phase II. Caustic
solution in Phase II exchanges CO2 and forms a precipitate at the bottom, finally yielding puri-
fied biogas. Biogas was circulated inside the caustic solution using an external compressor. The
raw and purified biogas concentrations at the inlet and outlet from the scrubber units were
monitored using an Infra Red Gas analyzer. Pressure and temperature inside the scrubber was
Figure 1. Basic layout of experimental setup for electricity generation from biogas.
624

Figure 2. Biogas purification systems.
above atmospheric condition. Sedimentation contents from the caustic scrubbing deposition
had accumulated at the bottom side and regular pH measurements were made using a digital
pH meter. All chemical reactions were carried out multiple times for precision.
A pilot study was carried out with various combinations of biodegradable waste that had
been generated at the university canteen, KVASU, Mannuthy, Kerala. Co-digestion of the veg-
etable leftovers with cow dung, maintained at a constant temperature (37°C), gave the maximum
biogas yield. Hence, this combination was used when evaluating the efficiency of scrubbing units.
2.3 Chemical reactions

Chemical reactions taking place in the scrubbers are responsible for the removal of unwanted
gases from the biogas. The reactions taking place in both the scrubbers are as follows.
2.4 Chemical reactions in Phase I

Phase I is filled with limestone and mild steel scraps. Iron oxide in the mild steel removes H2S,
whereas calcium oxide in limestone removes CO2 and NH3, as shown in Equations 1, 2 and 3.
H2S removal: 2Fe2O3 + 6H2S → 2Fe2S3 + 6H2O (1)

2Fe2S3 + 3O2 → 2Fe2O3 + 6S (2)
CO2 removal: CaO + CO2 → CaCO3 (3)
The calcium carbonate, iron sulfide, calcium and nitrogen formed during these reactions
do not have much effect on the performance of the biogas and can be removed along with
scrubbing material at regular intervals. Thus, the biogas coming out of scrubber 1 has a lesser
amount of CO2, H2S and NH3 than it does at the inlet [Katare et al. 2016].
2.5 Chemical reactions in Phase II

Chemicals like NaOH, KOH and Ca(OH)2 were used for the removal of CO2. Sodium
hydroxide is also a caustic solvent, which his categorized as being a strong alkaline. NaOH
ionizes into Na+ and OH- in water. CO2 will be chemically absorbed by NaOH, as shown in
Equation 4. Potassium hydroxide (KOH) is an abundant caustic solvent, used for up-gradation
of biogas [Cebula, et al. 2009]. KOH reacts with CO2, as shown in Equation 5. There was no
difference between the chemical absorption of CO2 using KOH and NaOH. KOH is more
expensive than NaOH, but it is advantageous due to the K2CO3 formed, which has several
industrial applications [Ghatak et al 2016].
625

2NaOH (aq) + CO2(g) → Na2CO3(aq) + H2O(l) (4)
2KOH (aq) + CO2(g) → K2CO3(aq) + H2O(l) (5)
The third caustic solvent utilized in the study to absorb CO2 is calcium hydroxide (Ca(OH)2).
The reaction of Ca(OH)2 with CO2 is given by Equation 6.
Ca(OH)2(aq) + CO2(g) → CaCO3(aq) + H2O(l) (6)
In all the above reactions, it can be seen that CO2 is absorbed by chemicals used in the
scrubber. Thus, the biogas coming out of scrubber II has a lesser amount of CO2 than it does
at the inlet [Leonzio et al. 2016].
The raw biogas is allowed to pass through the scrubbing system. Initial purification was
done using a single column packed unit for the removal of CO2, H2S and NH3, as shown in
[Fig. 2]. The composition of raw biogas (i.e. at the inlet and outlet of the Phase I scrubber) is
as shown in [Table 1] and [Fig. 3].
Scrubbing at Phase I reported a 5% increase in methane and a CO2 removal of 5.8%.
A removal rate of 43% and 37% was observed for H2S and NH3 respectively. Experimental
data from Phase I reported better H2S removal, as compared with [Katare, et al. 2016].
A further purification process was carried out using saturated caustic solution of NaOH,
KOH and Ca(OH)2. The results for CO2 removal are shown in Table 2 and Figure 3. The
caustic absorption process varied with time duration (Fig. 4).
It was found that different caustic solutions gave dissimilar CO2 levels at the scrubber
outlet. A nonlinear graph plotted clearly indicated a variation in the CO2 removal rate, along
with the type of caustic solution, and was dependent on time. Carbon dioxide sequestration
Table 1. Biogas composition at inlet and outlet

of Phase 1 scrubber.
Gas Inlet Outlet
CH4 53.1% 58.1%

CO2 45.8% 40.3%
H2S 721 ppm 315 ppm
NH3 75 ppm 28 ppm
Figure 3. Phase I scrubber performance.
626

using different caustics is represented in the Fig. 4. The unique design to delay biogas from
bubbling inside the apparatus is the key object for efficient purification. Therefore, the final
results reported were better than those of [Ghatak, et al. 2016].
Carbon dioxide was removed at a rate of 34.9% for KOH, followed by NaOH and Ca(OH)2
at a rate of 34.1% and 33.9% respectively, in a time duration of three minutes. All the data in
[Table 2] was greater than reported by [Cebula, J. et al. 2009]. It has undoubtedly been indi-
cated that the quality of biogas increased with the use of caustic scrubbing solution. Among
all caustic chemicals, KOH reported the maximum efficiency of CO2 removal rate (Fig. 5).
All caustic solution stability used to purify biogas is clearly plotted in Figure 6 with respect
to time, and it was found that KOH had maintained pH stability for a longer time period.
Ca(OH)2 solution is very unstable, and the pH dropped rapidly in a time period of 15 minutes.
Table 2. CO2 removal with caustic solutions for

3 minutes.
Biogas
content Ca(OH)2 NaOH KOH
CH4 (%) 94.3 94.6 96.1

CO2 (%) 6.4 6.2 5.4
CO2 removal 33.9 34.1 34.9
rate (%)
Figure 4. CO2 removal rate with time.
Figure 5. Variation of composition of CO2 and CH4 with different chemical solution.
627

Figure 6. Variation of pH with time.
It can be seen from the findings that saturated caustic solutions are highly suitable for the
continuous purification process. Final results were in accordance with the reported literature.
Results obtained were inferior to those from [Tippayawong & Thanompongchart, 2010]. The
maximum H2S and CO2 removal rate was obtained when compared to the caustic scrubbing
system developed by [Ürsein et al. 2015]. Enhanced H2S absorption was observed, compared
to that stated by [Miltner, M., Makaruk et al. 2012]. Efforts were made to create purified
methane enriched gas for a sustained period. It has been found that absorption capacity of
caustic solution was dropping within a short time period. Hence, it is necessary to replace
the caustic solution is necessary in order to uphold the chemicals at saturation point. Regu-
lar replacement results in concentration instability and the major drawback in consuming
caustic solvents was that they are very problematic to recycle. Although they are compara-
tively cheaper, a huge quantity of chemicals is essential to fulfill the purification process and
to overcome the drop in engine efficiency accounted for by the presence of CO2 in biogas.
Amount utilized in this present study is high for making purified biogas. So, future work is
mandatory to reduce the capital cost in biogas enrichment and its applications.
4 CONCLUSION
Sequestration of CO2, NH3 and H2S from biogas by a multistage scrubber was studied.
Enhanced H2S scrubbing effectiveness has been gained in the Phase 1 scrubber. Meanwhile,
the effectiveness was found to be dropping with time. NaOH, KOH and Ca(OH)2 were used
in the current study and their absorption behaviors were observed. The absorption character-
istics of all caustic solvents indicated similar results. Chemical absorption by caustic solution
was found to be an effective technique for small process time but their absorption capability
weakened quickly with time. Chemical absorption with caustic solution is not advisable as
an alternative for biogas quality upgrade, due to its limitation for being reused. Still, caustic
scrubbing techniques are considered as a low-cost sequestration method for sub-continental
conditions. In addition, capturing CO2 into solid phase, instead of it being released into the
atmosphere, makes the projected enrichment process more environmentally friendly.
REFERENCES
Abdeen, F.R., Mel, M., Jami, M.S., Ihsan, S.I. & Ismail, A.F. (2016). A review of chemical absorption
of carbon dioxide for biogas upgrading. Chinese Journal of Chemical Engineering, 24(6), 693–702.
Cebula, J. (2009). Biogas purification by sorption techniques. Architecture Civil Engineering Environ-
ment Journal, 2, 95–103.
628

Ghatak, M.D. & Mahanta, P. (n.d.) Biogas Purification using Chemical Absorption (2016).
Kadam, R. & Panwar, N.L. (2017). Recent advancement in biogas enrichment and its applications.
Renewable and Sustainable Energy Reviews, 73, 892–903.
Kapdi, S.S., Vijay, V.K., Rajesh, S.K. & Prasad, R. (2005). Biogas scrubbing, compression and storage:
Perspective and prospectus in Indian context. Renewable Energy, 30(8), 1195–1202.
Katare, S. (2016). Biogas purification & enrichment through stepped scrubbing. International Journal of
Engineering Sciences & Research Technology, 1(5), 795–799.
Leonzio, G. (2016). Upgrading of biogas to bio-methane with chemical absorption process: Simulation
and environmental impact. Journal of Cleaner Production, 131, 364–375.
Miltner, M., Makaruk, A., Krischan, J. & Harasek, M. (2012). Chemical-Oxidative scrubbing for
the removal of hydrogen sulphide from raw biogas: Potentials and economics. Water Science and
Technology, 66(6), 1354–1360.
Osorio, F. & Torres, J.C. (2009). Biogas purification from anaerobic digestion in a wastewater treatment
plant for biofuel production. Renewable Energy, 34(10), 2164–2171.
Privalova, E., Rasi, S., Mäki-Arvela, P., Eränen, K., Rintala, J., Murzin, D.Y. & Mikkola, J.P. (2013).
CO2 capture from biogas: Absorbent selection. RSC Advances, 3(9), 2979–2994.
Sanaei-Moghadam, A., Abbaspour-Fard, M.H., Aghel, H., Aghkhani, M.H. & Abedini-Torghabeh,
J. (2014). Enhancement of biogas production by co-digestion of potato pulp with cow manure in a
CSTR system. Applied Biochemistry and Biotechnology, 173(7), 1858–1869.
Scano, E.A., Asquer, C., Pistis, A., Ortu, L., Demontis, V. & Cocco, D. (2014). Biogas from anaerobic
digestion of fruit and vegetable wastes: Experimental results on pilot-scale and preliminary
performance evaluation of a full-scale power plant. Energy Conversion and Management, 77, 22–30.
Surata, I.W., Nindhia, T.G.T., Atmika, I.K.A., Negara, D.N.K.P. & Putra, I.W.E.P. (2014). Simple
Conversion method from gasoline to biogas fueled small engine to powered electric generator. Energy
Procedia, 52, 626–632.
Tippayawong, N. & Thanompongchart, P. (2010). Biogas quality upgrade by simultaneous removal of
CO2 and H2S in a packed column reactor. Energy, 35(12), 4531–4535.
Üresin, E., Saraç, H.İ., Sarıoğlan, A., Ay, Ş. & Akgün, F. (2015). An experimental study for H2S and CO2
removal via caustic scrubbing system. Process Safety and Environmental Protection, 94, 196–202.
629

Development of ceramic membrane for microfiltration

application in biotechnology field
C. Mohit Kumar & D. Vasanth

Department of Biotechnology, National Institute of Technology, Raipur, India
ABSTRACT: A new composition of precursors was identified using china clay, quartz and
calcium carbonate to fabricate the microfiltration membrane. The membrane was fabricated
by pressing method and sintered at 1000°C. Various characteristics of membrane such
as porosity, average pore size, water permeability and chemical resistance were evaluated.
Energy Dispersive X-ray analysis (EDX) was conducted to identify the elements present in
the membrane. The porosity, water permeability and pore size of membrane are found to be
37%, 2.88 × 10−3 L/m2.h.Pa and 555 nm respectively. Corrosion resistance test indicates that
the membrane can be subjected to acid and alkali based cleaning procedure.
1 INTRODUCTION
In recent years, the preparation of clay based inexpensive ceramic membrane is getting
significant attention due to its cost benefits. The ceramic membrane could be deployed
to highly corrosive medium and high pressure applications. Numerous articles have been
published for the preparation of clay based ceramic membranes. Nandi et al. (2009)
formulated new composition of raw materials using kaolin, quartz, sodium carbonate,
calcium carbonate, boric acid and sodium metasilicate to synthesize circular membrane. The
prepared membrane was deployed for the purification of oil-water emulsions. Abbasi et al.
(2010) utilized kaolin, clay and α-alumina to synthesize mullite and mullite–alumina based
ceramic membranes for the separation of oil emulsions. Similarly, the ceramic membrane was
manufactured using mixture of kaolin, pyrophyllite, feldspar, quartz, calcium carbonate, ball
clay and titanium dioxide by uniaxial compaction method. The membrane was performed
well for the treatment of oil-water emulsions (Monash & Pugazhenthi 2011). Using perlite
materials, Al-harbi et al. (2016) prepared the super-hydrophilic membrane (mean pore size
of 16 µm) for the wastewater treatment applications. In another work, dairy wastewater was
treated using novel tubular ceramic membrane. The membrane was prepared using mixture
of naturally available clays (Kumar et al. 2016). Jeong et al. (2017) used the pyrophyllite and
alumina to prepare composite ceramic membrane (pore size of 0.15 µm). The membrane
performance was investigated for the treatment of low-strength domestic wastewater.
The detailed investigation of above literatures indicates that the clay based membranes
were mainly deployed for wastewater treatment applications. To best of our knowledge, the
applicability of clay based ceramic membrane in biotechnological field is less studied. In this
context, the applicability of clay based ceramic membrane for biotechnological field needs to
be investigated. Such research would be useful to understand upon the suitability of ceramic
membrane for biotechnological applications.
This article addresses the preparation of ceramic membrane using china clay, quartz and
calcium carbonate. Primary characteristics such as porosity, average pore size, water permea-
bility was evaluated. Corrosion resistance test was conducted to identify the suitable cleaning
631

procedures and its stability in highly corrosive medium. Eventually, the production cost of
membrane was determined to compare the cost of other reported membrane in literatures.
2 EXPERIMENTAL
2.1 Precursors
The raw materials namely china clay, quartz and calcium carbonate were used to develop
ceramic membrane and its composition is presented in Table 1. China clay and quartz were
purchased from Royalty minerals, Mumbai, India. The calcium carbonate was procured from
Loba Chemie, Ltd. The materials were used without any pretreatment.
2.2 Synthesis of membrane

Using uni-axial compaction method, the ceramic membrane was prepared with newly identi-
fied precursor composition. Different sequential steps involve in the preparation of mem-
branes is illustrated in Fig. 1. Initially, the raw materials were uniformly mixed with 4 ml of
2% polyvinyl alcohol (binder) using mortar and pestle. The necessary amount of powder
was then poured in SS mould and pressed at higher pressure (50 MPa) using hydraulic press
(Guru Ramdas Machine Works, Raipur, India) to cast the membrane (52 mm diameter and
5 mm thickness). The obtained membrane was dried at 100°C to remove moisture present
in the membrane using hot air oven (Pooja Scientific Instruments, India). Subsequently, the
membrane was sintered at 1000°C for 5 h in a muffle furnace (Nanotec, Chennai, India).
After that, the membrane was polished at both sides using abrasive sheet (No. 220) and
washed in ultrasonicator (PCI Analytics, India). Eventually, the membrane was dried at
100°C for 12 and subjected to the various characterizations. The photograph of prepared
membrane is shown in Fig. 2.
Table 1. Composition of raw materials.
Materials Quantity (gm)
China clay 50
Quartz 25
Calcium carbonate 25
Figure 1. Flow chart for membrane preparation.
632

Figure 2. Image of prepared membrane. Figure 3. Experimental setup.
2.3 Characterization
The primary characteristics of membrane such as porosity, water permeability, average
pore size, chemical resistance were evaluated. The porosity of membrane was evaluated
by Archemede’s principle using water as wetting liquid (Nandi et al. 2009). The water flux
(J, L/m2h) of membrane was measured using indigenous continuous dead-end filtration setup
(Fig. 3). The flux was measured at different applied pressure (69–345 kPa) at room tem-
perature (25°C). This involves the measurement of permeate volume at an interval of 5 min
during the total run time of 25 min. The flux was calculated using the relation.
J = Q/A × t. (1)
where, Q is the volume of permeate collected, t is the time and A is effective membrane area
(m2) for permeation.
The water permeability (Lh) and average pore size (rl) of the membrane was determined
from water flux data according to the following expression
ε r 2 ∆P
Jv = = Lh ∆P (2)
8 µl
where Jv (L/m2h) is the water flux through the membrane, ∆P (kPa) is the trans-membrane
pressure drop across the membrane, µ is the viscosity of water, l is pore length, ε is the
porosity of the membrane.
The corrosion resistance of membrane was tested in acid and alkali solutions individually at
different pH levels (1–14) using HCl and NaOH. To do so, the membrane was kept in contact
with acid and alkali solutions for seven consecutive days at room temperature. After that, the
weight loss of membrane was measured that characterizes the corrosion resistance of membrane.
In addition, EDX was performed to confirm the elements present in the membranes.
3.1 Porosity
The porosity was determined by Archimedes principle using water as wetting liquid. Generally,
the pore size of membrane is depending upon the porosity. In this work, the porosity of
membrane is found to be 37%. In this context, it can be pointed out that the obtained porosity
(37%) is comparable even higher than the porosity of cordierite (36%), mullitte (32%) and
kaolin (36%) (Dong et al. 2007, Abbasi et al. 2010, Monash & Pugazhenthi 2011). Thus, it is
inferred that the formulated composition of raw material provides higher porosity.
633

3.2 Water flux and permeability
The water flux was measured at different trans-membrane pressure (69–345 kPa) for total run
time of 25 min at time interval of 5 min. The time versus flux at different applied pressure
is presented in Fig. 4. It can be observed that the flux is constant over entire studied time.
It indicates that there is no any resistance for pure water flow. Further, it is noted that water
flux increases linearly with increasing pressure (Fig. 5). Using the water flux data, water
permeability of membrane is evaluated to be 2.88 × 10−3 (L/m2hPa).
3.3 Average pore size

Using the water flux data, the average pore diameter of membrane is evaluated to be 555 nm.
In this context, it can be pointed out that the reported pore size (555 nm) is significantly
lesser than the average pore size obtained with different precursors namely Moroccan clay
(10.75 µm), cordierite (8.66 µm) and apatite (5 µm) (Saffaj et al. 2006, Dong et al. 2007,
Masmoudi et al. 2007). The obtained membrane with lesser pore size can be applied for
various biotechnology applications.
3.4 Corrosion resistance

Corrosion resistance test was performed to identify the suitable chemical cleaning procedure
and its application in highly corrosive medium. In view of this, the membrane was subjected
to treat in acid and alkali solutions at different pH range (1–14). It is observed that there is no
significant weight loss in both acid and alkali. This result shows the membrane have excellent
corrosive resistivity that can be used for any separation application whose pH range between 1
and 14. Further, the membrane can be subjected to acid and alkali based cleaning procedure.
Figure 4. Time vs Flux. Figure 5. Pressure vs Flux.
Figure 6. EDX analysis of membrane.
634

Table 2. EDX analysis of membrane.
Elements Weight (%)
Carbon 16.24
Oxygen 58.01
Aluminum 5.95
Silicon 14.85
Calcium 4.95
Table 3. Determination of cost.
Raw materials used

Unit for preparation of Cost of one
price/kg one membrane membrane
Raw materials (Rs.) (gm) (Rs)
China clay 11.44 10 0.11

Quartz 11.40 5 0.06
Calcium 297 5 1.48
carbonate
Total 1.65
Table 4. Summary of characterization results.
Properties Values
Porosity (%) 37
Water permeability (L/m2 h Pa) 2.88 × 10-3
Pore size (nm) 555
3.5 EDX analysis

Fig. 6 illustrates the EDX analysis of membrane. The EDX graph was obtained to iden-
tify the elemental composition of membrane. It is analyzed that the main elements that are
present in the membrane are carbon, oxygen, aluminum, silicon and calcium and their rel-
evant compositions are present in Table 2.
3.6 Cost estimation

The research in the preparation of low cost ceramic membrane is intensified worldwide.
Generally, the cost of any membrane is depending upon the raw material cost and fabrica-
tion method. Therefore, the selection of inexpensive raw materials and simple fabrication
technique is vital. Based on the raw material cost, the cost of membrane is estimated to be
Rs. 1.65 (5 mm thickness and 52 mm diameter). The details of cost estimation are shown in
Table 3. Moreover, the overall characterization results is presented in Table 4.
4 CONCLUSIONS
A new raw material composition was identified for the preparation of ceramic microfiltra-
tion membrane. The porosity, water permeability and pore size of membrane are found to
be 37%, 2.88 × 10−3 L/m2.h.Pa and 555 nm, respectively. Corrosion resistance test indicates
the membrane can be subjected to acid and alkali based cleaning procedure. Henceforth, it
is concluded that the fabricated ceramic membrane possesses very small pore size which is
suitable for various biotechnology applications.
635

ACKNOWLEDGEMENT
The authors wish to express their sincere thanks to DST-SERB for the financial support.
REFERENCES
Abbasi, M. Mirfendereski, M. Nikbakht, M. Golshenas, M. & Mohammadi, T. 2010. Performance

study of mullite and mullite–alumina ceramic MF membranes for oily wastewaters treatment.
Desalination 259: 169–178.
Al-Harbi, O.A. Mujtaba, K.M. & Ozgur, C. 2016. Designing of a low cost super-hydrophilic membrane
for wastewater treatment. Materials and Design 96: 296–303.
Dong, Y. Feng, X. Dong, D. Wang, S. Yang, J. Gao, J. Liu X. & Meng, G. 2007. Elaboration and
chemical corrosion resistance of tubular macro-porous cordierite ceramic membrane supports.
Journal of Membrane Science 304: 65–75.
Jeong, Y. Lee, S. Hong, S. & Park, C. 2017. Preparation, characterization and application of low-cost
pyrophyllite alumina composite ceramic membranes for treating low-strength domestic wastewater.
Journal of Membrane Science 536: 108–115.
Kumar, R.V. Goswami, L. Pakshirajan, K. & Pugazhenthi, G. 2016. Dairy wastewater treatment using
a novel low cost tubular ceramic membrane and membrane fouling mechanism using pore blocking
models. Journal of Water Process Engineering 13: 168–175.
Masmoudi, S., Larbot, A., El Feki, H. & Amar, R.B. 2007. Elaboration and characterisation of apatite
based mineral supports for microfiltration and ultrafiltration membranes. Ceramics International 33:
337–344.
Monash, P. & Pugazhenthi, G. 2011. Effect of TiO2 addition on the fabrication of ceramic membrane
supports: a study on the separation of oil droplets and bovine serum albumin (BSA) from its solu-
tion. Desalination 279: 104–114.
Nandi, B.K. Uppaluri, R. Purkait, M.K. 2009. Treatment of oily wastewater using low cost ceramic
membrane: flux decline mechanism and economic feasibility. Separation Science and Technology 44:
2840–2869.
Saffaj, N. Persin, M. Younsi, S.A. Albizane, A. Cretin M. & Larbot, A. 2006. Elaboration and charac-
terization of microfiltration and ultrafiltration membranes deposited on raw support prepared from
natural moroccan clay: application to filtration of solution containing dyes and salts. Applied Clay
Science 31: 110–119.
636

Studies on carbon felt/phenolic composites as light weight

advanced ablative TPS for launch vehicle programmes
C.Y. Lincy
Department of Chemical Engineering, GEC Thrissur, Kerala, India
V. Sekkar, Vijendra Kumar & Ancy Smitha Alex

Chemical Systems Group, PCM Entity, Vikram Sarabhai Space Centre, Thiruvananthapuram, Kerala, India
Saurabh Sahadev
Department of Chemical Engineering, GEC Thrissur, Kerala, India
ABSTRACT: Carbon felt based phenolic composites were prepared at various densities viz.
0.3, 0.4, 0.5, 0.6 and 0.7 g/cc. The mechanical and thermal characteristics of the composites
were evaluated adopting standard test procedures. With the change in density from 0.3 to
0.7 g/cc, flexural strength enhances from 5.2 to 32 MPa, the compressive strength increases
from 1.1 to 6.9 MPa, tensile strength is pushed up to 225 from 45 KSC while resilience improves
from 1.5 to 1.8 kJ/m2. Over the change of density, thermal conductivity of the composites
marginally increases from 0.1 to 0.2 W/mK. Char yield for the composites as determined by
thermogravimetric analysis was about 60% up to 900°C. Ablative characteristics were deter-
mined through plasma arc jet simulated test procedures at an energy flux of 50 W/cm2 and time
duration of 10 seconds; the heat of ablation was 2044 cal/g, and composite with 0.7 g/cc density
had survived the erosion test with an erosion rate of 0.002 mm/s and a mass loss of 0.005 g.
1 INTRODUCTION
In a launch vehicle momentum for the rocket is gained through conversion of chemical energy
to mechanical energy. In a typical propulsion system, fuel and oxidizer undergo combustion
thereby generating low molecular weight gases at high temperatures (3500 K) and pressure
(200 bar). When this hot high pressure gas is expanded through nozzle, thrust is created and
the rocket experiences forward motion. For most of inner wall portions of the nozzle tem-
perature would be in the range 700–1200 K. All known structural materials cannot survive
under such severe erosive and thermal shock conditions. Thus the metals form nozzle needs
protection from very high speed extremely hot gas streams. Traditionally, highly dense silica/
phenolic and carbon/phenolic composites (1.8 g/cc) are employed as nozzle liners. Ablation
lining thickness is about 8–12 mm. It is envisaged that the fully dense liner is replaced with a
porous, low dense and light weight material with a thick anti erosion coating, a huge reduc-
tion in weight of the nozzle with improved performance as the porous material would have
a much lower thermal conductivity than the fully dense counterparts. In view of the above,
carbon felt impregnated phenolic composites are investigated for their role as light weight
ablative liners. Carbon felt-based ablators have several advantages over classical fully dense
ablative liner materials. Most importantly, they reduce the limited strain response of large
rigid substrates. Carbon felt materials are known for their benign insulating property, uni-
form bulk density and better shape retention properties with macro and micro communica-
tion channels among the cells allowing efficient resin infiltration. Enabling manufacturing in
larger sizes, felt based substrates reduce the number of independent parts mitigating the need
of gap fillers. They also offer improved robustness in absorbing loads and deflections, and
637

they allow shaping the substrate around complex geometries with the possibility of maintain-
ing a uniform and low thermal conductivity in the system. In this study, the evaluation of
mechanical as well as thermal performance of the composites are carried out as the prelimi-
nary screening exercise before such low density composites are considered for the intended
applications. Because of the highly porous nature, it enhances efficient infiltration of the
resin and accounts for its uniform properties.
2.1 Materials required

The materials used for the preparation of carbon felt based phenolic composites are phenolic
resin PF108, carbon felt and acetone as the solvent. Phenol formaldehyde resin (PF 108) was
produced at PFC, VSSC, Trivandrum. Carbon felt was procured from Mersens India Limited,
Bangalore. Acetone was procured from Merck Specialities India Limited, Bangalore.
2.2 Preparation of carbon felt/phenolic composites

The carbon felts were cut to the required size. Required amount of PF 108 resin was dissolved
in acetone and the solution was completely absorbed over carbon felts. The solvent was driven
Table 1. Calculation of weight of PF108 for preparation of com-

posites of each densities.
Required Felt Felt Weight of

Composite density volume weight PF108
name (g/cc) (cc) (g) (g)
A 0.3 402 49.72 98.42

B 0.4 393 50.39 161.44
C 0.5 395 51.5 204.53
D 0.6 393 52.5 254.58
E 0.7 391 53.6 305.65
Figure 1. Carbon felt/phenolic composites after preparation.
638

off by drying the soaked felt at 50°C for 18 hrs. Curing was undertaken at 120°C for 1 hr and
at 180°C for 4 hrs. In a similar way felt composites with densities ranging from 0.3 to 0.7 g/cc
were prepared. The cured composites were characterized for their mechanical and thermal
properties, such as resilience, tensile strength, compressive strength, flexural strength and ther-
mal conductivity, adopting procedures conforming to relevant ASTM standards.
3.1 Mechanical properties

Carbon felt based phenolic composites with densities varying between 0.3 and 0.7 g/cc
were prepared and characterized for both mechanical and thermal characteristics relevant
to their application as low density ablative nozzle liner applications. Phenolic Carbon felt
composites may be better suited to intended applications chiefly due to spectacular specific
strength capabilities and very low thermal conductivities. PF resin during the curing proc-
ess looses as much as 25% by weight which is considered in calculating the resin intake. In
addition to that, felt offers easy handling and structural tailor ability for the composites.
Phenolic composites are best suited for ablation. Phenolic composites are not inflammable
but undergo endothermic decomposition which effectively remove heat from the system and
cool it. Further, carbon derived from ablation of phenolic composites belongs to the class
of glassy carbons which are known for their extremely low thermal conductivity at fairly
high temperatures.
Carbon felt/phenolic composites thus satisfy three important requirements for an abla-
tive TPS, viz.: good mechanical characteristics, low thermal conductivity and low density.
In particular, the density is characterized by a lower limit dictated by the necessity of a low
recession rate, and an upper limit dictated by the light weight requirements of all the aero-
space parts. Carbon felt based phenolic with low densities during ablative process contribute
to the formation of a porous char skeleton where the gases can flow through as the degrada-
tion reaction evolves. Thus it helps in ablative cooling and limiting heat transfer and serves
as good TPS material.
In general, all mechanical characteristics increase with density. Structural strength compo-
nent comes from the phenolic matrix. However, the fibrous portion of the felt thoroughly and
uniformly reinforce the phenolic matrix, which contributes surprisingly very high strength for
the densities encountered in this study. Compressive strength, flexural strength and tensile
strength increase linearly with density while impact strength has non linear dependence on
density. Impact strength obtained in this study is fairly high for the densities of the composites.
Materials at the similar densities without felt would be significantly brittle and would asso-
ciate with much lower strength figures. It is evident that the resin is completely responsible.
Figure 2. Variation of tensile strength with density.
639

Figure 3. Variation of flexural strength with density.
Figure 4. Variation of compressive strength with density.
Figure 5. Variation of impact strength with density.
With a density variation from 0.3 g/cc to 0.7 g/cc, the flexural strength changed from 5.2 to
32 MPa, the resilience changed from 1.5 to 2.5 kJ/m2, compressive strength changed from
1.1 to 6.9 MPa and tensile strength changed from 4 to 22 MPa. The compressive and tensile
strength of the composite with 0.7 g/cc density is highly remarkable when compared to the
conventional TPS materials for the given density.
640

3.2 Thermal conductivity
Thermal conductivity does not get altered much during the density range. All the composites
have thermal conductivity between 0.1 and 0.2 W/mK. This is a very favorable aspect as the
felt composites have a wonderful combination of remarkable mechanical strength, low density
and very low thermal conductivity. Significantly larger pore spaces are reason for extraordinary
lower thermal conductivity. The thermal conductivity values are much lower than the typical
high density counterparts. Thus, combination of low thermal conductivity with fairly higher
mechanical strength parameters makes them ideal candidates for ablative TPS applications.
3.3 Thermo gravimetric analysis

The thermo gravimetric analysis conducted on the sample of density 0.7 g/cc and weight
5.92 mg showed that the decomposition of the sample occurs in the temperature range
between 300°C to 600°C.
The TGA curve is showed in Figure 7. It is evident from the analysis that the residue
remained at 900°C is 59.64%. From the different analyses conducted on the specimen, it is
obvious that the carbon felt/phenolic composites show superior performance. In order to
investigate more on its ablation characteristics, a plasma arc jet test was also conducted which
is explained in the following section.
Figure 6. Variation of thermal conductivity with density.
Figure 7. TGA curve for carbon felt/phenolic composites.
641

Figure 8. Specimen after ablation test.
3.4 Ablation test

In order to screen a material for its aptitude for ablation application, it needs to be subjected
to plasma arc jet test, by which important ablation characteristics, such as heat of abla-
tion, erosion rate and mass loss are evaluated. Plasma arc jet test was performed adopting
the procedure conforming to test standard ASTM E258. The samples are tested at a heat
flux of 50 W/cm2 for 10 seconds; the test was conducted in environment of argon Ablative
properties were evaluated for carbon felt reinforced phenolic with varying densities viz. 0.3
and 0.7 g/cc. However, the composite with 0.3 g/cc could not tolerate the thermo chemical
and mechanical stresses and it failed during the ablative evaluation. Composite at 0.7 g/cc
survived the erosion test with an erosion rate of 0.002 mm/s and a mass loss of 0.005 g. For a
heat flux of 50 W/cm2, the heat of ablation was found to be 2044 cal/g. The values obtained
are very promising qualifies the composite as a possible TPS material. Further investigations
at higher heat flux are necessary for the complete analysis of its ablation property.
4 CONCLUSION
The carbon felt phenolic composites exhibit superior mechanical properties. Thermal con-
ductivity values are very small. Low density felt composites are thus attractive materials for
nozzle liner end uses.
REFERENCES
de Almeida, L., Cunha, F., Batista N., Roccol, J., Iha K. & Botelho E. 2014. Processing and charac-
terization of ablative composites used in rocket motors. Reinforced Plastics and Composites 33(16):
1474–1484.
Francesco Panerai, Joseph C. Ferguson Jean Lachaud, Alexandre Martin, Matthew J. Gasch & Nagi N.
Mansour. 2016. Micro-tomography based analysis of thermal conductivity, diffusivity and oxidation
behavior of rigid and flexible fibrous insulators. Heat and Mass Transfer. 108(2017) 801–811.
Ganeshram, V. & Achudhan, M. 2013. Synthesis and characterization of phenol formaldehyde resin as
a binder used for coated abrasives. Science and Technology. (6S), 0974-6846.
Varadarajulu & Rama Dev, R. 2008. Flexural properties of ridge gourd/phenolic composites and glass/
ridge gourd/phenolic hybrid composites. Composite materials. 42, 6/2008.
Vineta, S. & Gordana B. 2009. Composite material based on an ablative phenolic resin and carbon fib-
ers. Serbian chemical society. 74(4): 441–453.
642

Desalination studies on two single-sloped solar stills with heat-

absorbing materials as coating material on aluminum basin
Pativada Suman, Deepa Meghavathu & Sridevi Veluru

Department of Chemical Engineering, Andhra University College of Engineering,
Visakhapatnam, Andhra Pradesh, India
Zakir Hussain
Department of Chemical Engineering, Rajiv Gandhi Institute of Petroleum Technology, Jais,
Amethi, Uttar Pradesh, India
ABSTRACT: Fresh water is the component required for the survival of many living
organisms. Though 71% of the earth’s surface is covered by water, only 3.5% is available for
human needs, with the remaining 96.5% in the form of oceans. Large quantities of water are
consumed by industries due to rapid industrialization, thus polluting the fresh water and
further causing its scarcity. To meet the demand, solar energy can be used to convert seawater
into fresh water in a solar distillation unit using a solar still. The solar still should possess
high absorptivity so that it captures the maximum solar energy and converts the seawater
to fresh water. In this paper, the effect of different materials such as black paint, paraffin
wax, coal bed, sand and ceramic packings, as coating materials in various combinations to
the base aluminum basin, was studied in a detailed way. Energy balance and heat transfer
coefficients were estimated for all the cases. Further, the overall efficiency was also estimated
for all the cases and compared. The black paint coated basin (base case) has shown the least
efficiency (42.19%) among all considered cases, and the basin coated with the combination of
black paint, sand and ceramics has shown the highest efficiency (65.06%).
Keywords: desalination, solar stills, heat-absorbing, energy balance, heat transfer coefficient
1 INTRODUCTION
Water is the main source of survival for many living organisms, and humans especially rely
on fresh water for consumption and household purposes. Around 71% of the earth’s surface
is covered with water but only 3.5% is available in the form of fresh water, while 1.7% is in
the form of glaciers, 1.7% is in the form of groundwater and 0.1% is in the form of rivers
and lakes (Sethi & Dwivedi, 2013). Not only humans but for all the life forms, water is an
essential commodity for their survival. The fresh water availability is less but the demand
for it is increasing day by day at a rapid rate, due to the increase in the population and also
due to rapid industrialization. Industrial wastes are disposed directly into the available fresh
water sources, thus polluting them and creating a rapid decline in the availability of the fresh
water (Panchal, 2015). According to the WHO (World Health Organization, 2017), around
25% of the human population does not have the provision of safe drinking water. With the
population growing by 82 million every year, the need for safe drinking water is increasing
day by day and as many as a third of humans will face a shortage of water by 2025 (Omara
& Kabeel, 2014). To mitigate and overcome this fresh water problem, many water purifica-
tion techniques are available for the production of clean water. Solar distillation is one such
technique that converts seawater into potable water. Solar energy is abundantly available in
nature and can be successfully utilized in treating the saline water, thus creating a demand
for solar distillation that is increasing every year. The basic purpose of solar distillation is
643

Figure 1. Schematic diagram of solar desalination unit.
to provide fresh drinking water from seawater in order to meet the demand for fresh water
(Somanchi et al., 2015). The process of solar distillation occurs in two ways based on the
energy consumption, namely thermal and non-thermal processes. In the solar distillation unit,
there are two types of solar stills that are used based on the energy utilization, namely active
and passive solar stills. Among these, active solar stills need an energy input in the form of
pumps for the input of feed into the system, whereas passive systems do not require energy
input. Hence, passive solar stills are preferred over active as they are more cost-effective. There
are many designs for solar stills, such as a single-slope solar still (Krishna et al., 2016), double
slope solar still (Murugavel et al., 2008), wick-type still (Suneesh et al., 2016), spherical still
(Dhirman, 1988), vertical still (Boukar, 2004), and multi-effect still (Tanaka et al., 2009). The
single-slope solar still provides a better productivity in the winter whereas the double slope
provides better results in the summer (Yadav & Tiwari, 1987). The design and fabrication of
a single-slope solar still is much easier when compared to that of the other designs as this
can be made from locally available materials like wood and polyurethane. Further, the design
involves a low maintenance cost and skilled labor is not required (Gugulothu et al., 2015).
A single-slope solar still is considered in the present work. The inclination of the single-slope
solar still must be equal to the latitude of the experimental location, in order for the maximum
solar irradiance to fall on the still to get the maximum yield (Malaeb et al., 2014). The addition
of a sand bed to the conventional solar still gives a higher productivity than with the normal
conventional solar still; the efficiency is increased from 35% in a conventional solar still, to
49% in a sand-bed solar still with a sand bed of height 0.01 m above the normal conventional
solar still basin (Omara & Kabeel, 2014). The operating principle of the solar still involves the
evaporation of the pure water molecules from the saline water due to the impingement of the
solar energy onto the saline water, evaporating out the water molecules and leaving behind
the dissolved salts and impurities at the bottom of the solar still. The process of impingement
of solar energy occurs by the incidence of the sun’s radiation on the glazing material of the
still. This allows the heat-absorbing material to absorb the radiation and heats up the seawater
present, with the evaporated water condensing and being collected through a collecting chan-
nel (Aburideh et al., 2012). A schematic diagram showing the principle is shown in Figure 1.
A single-slope distillation unit is made up of wood for the external body of the box with
dimensions of 0.715 m in length, 0.415 m in breadth, the height of the smaller edge is
0.125 m, the height of the longer edge is 0.36 m, and with a thickness of 0.01 m. Inside the
box, there is 0.02 m thick insulation that is made up of thermocol sheets. The basin that holds
the seawater is made up of aluminum (length 0.64 m, breadth 0.33 m, and height of 0.085 m)
and is coated with black paint. The glazing material, which is the glass cover that covers the
top of the unit, is 0.004 m thick, with an inclination of 18° that results in a heat-absorbing
area of 0.3154 m2. A collector is attached on the glass cover at the lower end to collect the
condensate. To avoid the glass slipping down and to reduce the vapor losses, a rubber seal is
provided between the wooden box and the glass cover. The base experimental setup is shown
in Figure 2.
644

Figure 2. Solar distillation unit with black paint coated basin (base case).
2.1 Different parts of the solar distillation unit

2.1.1 Still basin
This is an important part of the solar distillation unit, which is used to hold the water to be
distilled. Materials with high absorptivity are preferred as basin materials, so that the maxi-
mum heat energy can be absorbed. Various materials can be used as basin materials, such as
leather sheet, GE silicon, aluminum, copper, paraffin wax, sand, coal and black paint. The
present study is carried out using a base case consisting of an aluminum basin coated with
black paint. Further, different materials such as coal bed, sand, paraffin wax and ceramic
packing are used as extra layers in different combinations.
2.1.2 Side walls

The side wall material for the construction of a box to house the solar still should possess
enough rigidity to hold the still, without any disturbance due to external forces such as wind.
Further, the side wall acts as an insulation for effective heat transfer that takes place from the
system to the surroundings. The material of the side wall should possess very low thermal
conductivity in order for it to act as a good insulation agent. Different materials that can be
utilized as side wall material are wood, RPF (Reinforced Plastic with Fiber), thermocol, and
concrete. In the present study, wood is considered as the side wall material and for better insu-
lation purposes, thermocol sheets (0.02 m thickness) are used inside the wooden box.
2.1.3 Top cover

The passage where irradiation occurs on the surface of the basin is the top cover. Also, it is the
surface where condensate collects. A suitable material is selected as a top cover, based on a few
features such as its transparency to solar radiation, its non-absorbency of water, and having
a clean and smooth surface. The materials that can be used are glass and polyethylene. In the
present study, the glass material of 4 mm thickness is selected as the top cover (size: 76 × 41.5 cm).
2.1.4 Channel
The channel is used to collect the condensate that has formed on the surface of the top glass
cover. The materials that can be used are PVC, galvanized steel, and RPF. In the present
study, PVC material of 1 inch diameter is used.

The single-slope solar distillation unit with an aluminum basin coated with black paint was
kept in the sunlight from 9:30 a.m. to 12:30 p.m. in the easterly direction, and from 12:30 p.m.
645

to 5:30 p.m. in the westerly direction. For paraffin wax and coal bed materials, experimentation
continued until 7:30 p.m. For every hour, the solar radiation was measured by a pyranometer.
The temperatures were measured at different locations: the ambient temperature (Ta), the top
glass cover outside temperature (Tgo), the top glass cover inside temperature (Tgin), the moist
temperature (Tm), the basin temperature (Tb) and the distillate temperature (Td), with the help
of thermometers. Further, the produced distillate was collected for every hour using a measur-
ing cylinder. The same experimental procedure is carried out for treating different capacities of
seawater (2, 2.5, 3 liters) and for different materials used as coatings on the aluminum basin in
different combinations.
2.3 Data analysis

The efficiency of a still (η) is calculated by using the following Equation:
m*latent heat of vaporization of water

η= ∗ 100 (1)
rg*Is
2.3.1 Energy balance

The energy balance for the distillation unit is given as:
IsAg/Ab = IsrgAg/Ab + qg,sAg/Ab + qh,gAg/Ab + qk,airAk,air/Ab
+ qk,lAk,l/Ab + qk,bAb/Ab + (mcwhsat,g) Ab (2)
where
Is Solar radiation intensity, W/m2;
Ag Area of glass surface, m2;
Ab Area of basin surface, m2;
rg Reflectivity of the glass cover for visible light;
Ak,air Circumferential area of solar still covered by inside moist air, m2;
Ak,l Circumferential area of solar still covered by seawater, m2;
mcw Mass velocity of condensed water, kg/m2.sec;
hsat,g Enthalpy of water at saturation temperature, kJ/kg.
Considering the heat transfer from the cover to the atmosphere by convection:
qh,g = hg(Tg - Ta), W/m2 (3)
where Tg is the glass temperature, Ta is the ambient temperature, and hg is the convective heat
transfer coefficient given by the following formula:
hg = 5.7+ 3.8 v, W/m2 (4)
where the forced convection coefficient dependent on the wind velocity, v (m/s) = 3.5 m/sec.
The radiative heat transfer from the glass cover to the atmospheric air is given by the
formula:
qg,s = eg Cs [(Tg/100)4 - (Tsky /100)4], W/m2 (5)
where
Glass emissivity 0.88;
Constant Cs 5.667 W/m2 K4;
Tsky Sky temperature (Ta-20°C).
The conductive heat transfer from the bottom to the atmosphere may be formulated as:
qk,b = kb (Tb - Ta) W/m2 (6)
646

where
(1/kb) = (1/hin) + (∑(δi/λi)) + (1/ha) m2 K/W (7)
hin = Convective heat transfer coefficient at seawater interface, W/m2.
∑(δi/λi) = (δg/λg) + (δb/λb) + (δw/λw) + (δth/λth) m2 K/W (8)
where
δg Thickness of glass;
λg Thermal conductivity of glass (0.96 W/m.K);
δb Thickness of basin;
λb Thermal conductivity of basin (205 W/m-K for aluminum, 1.6 W/m-K for black
paint, 2.05 W/m-K for sand, 0.25 W/m-K for paraffin wax, and 0.33 W/m-K for coal);
δw Thickness of wood;
λw Thermal conductivity of wood (0.17 W/m-K);
δth Thickness of thermocol (20 mm);
λth Thermal conductivity of thermocol (0.036 W/m-K);
ha Convective heat transfer coefficient at ambient temperature, W/m2.K;
Tb Temperature of the basin, K;
Ta Ambient temperature, K.
Considering the heat transfer from the circumferential area of the still by conduction.
From inside moist air to the atmosphere:
qk,air = km (Tr - Ta), W/m2 (9)
where
(1/km) = (1/hr) + (∑ (δi/λi)) + (1/ha) (10)
hr Convective heat transfer coefficient of moist air, W/m2

From liquid to the atmosphere,
qk,l = kl (Tb - Ta) W/m2 (11)
where
(1/kl) = (1/∞) + (∑(δi/λi)) + (1/ha)m2.K/W (12)
2.3.2 Heat transfer coefficients (W/m2-K)
 1

) ( )
  pw − pgi (Tw + 273)  3 
(
hwc = 0.884  Tw − Tgi +    (13)
 268.9 * 10 − pw  
3

 
 5144 
Pw = exp  25.317 − , J/K (14)
 Tw + 273 
 5144 
Pgi = exp  25.317 − , J/K (15)
 Tgi + 273 
Pw − Pgi
he = 16.273 * 10 −3 * hwc * (16)
Tw − Tgi
647

The whole experiment was carried out during the months of March and April 2017. The
effects of various materials as coating for the aluminum basin in different combinations were
studied. Further, the effect of treating capacity of the solar still on the overall efficiency,
energy balance and heat transfer coefficients were studied by varying the treating volumes of
the seawater by 2, 2.5, and 3 liters respectively. For all the cases, the overall efficiency, energy
utilized (%) and heat transfer coefficients were estimated.
The effect of sand, ceramic, paraffin wax and coal bed materials, along with black paint,
for varying treating capacities of water, is shown in Figures 3, 4, 5, 6, 7, 8 and 9.
As shown in Figures 3 to 7, the optimum exposed time was found to be 1:30 p.m. for the
materials, black paint (with and without ceramics), black paint (with and without sand),
black paint (with sand + ceramics), and black paint + paraffin wax (with and without sand),
as the maximum incidence of solar irradiance was observed at that time.
As shown in Figures 8 and 9, the optimum exposed time was found to be 12:30 p.m. in the
case of black paint + coal bed (with and without ceramics). The optimum exposed time was
Figure 3. Effect of ceramics on base case. Figure 4. Effect on sand on base case.
Figure 5. Effect of (sand + ceramics) on base

case. Figure 6. Effect of paraffin wax on base case.
Figure 7. Effect of sand on paraffin wax. Figure 8. Effect of coal bed on base case.
648

observed earlier when compared to other combination of materials as the coal is considered
to be a heat-absorbing material and black in color.
As shown in Figures 6 to 9, it was observed that the distillate production is continued till
7:30 p.m., even in the absence of the sunlight.
The overall efficiency was estimated for all the considered cases, as shown in Figure 10. The
highest efficiency of 65.06% at 3 liters of treating volume capacity of seawater was exhibited
by the solar still coated with (sand + ceramics), with the lowest (53.01%) being the base case.
The energy utilized during the process was calculated for all the considered case with 3 liters
treating volume capacity and the same is shown in Figure 11. The energy utilization was found
to be maximum in the case of solar still basin coated with paraffin wax + sand. The combi-
nation of paraffin wax along with sand bed was proved to have an improved heat-absorbing
capacity, with a minimum amount of energy losses compared to the other combinations.
The evaporative heat transfer coefficient of water to the glass surface was calculated for all
the cases and compared. Figure 12 shows the comparison of the heat transfer coefficients for
the treating volume of 3 liters.
The evaporative heat transfer coefficient was observed to be the highest (2964.83 W/m2 K)
in the case of a solar still basin coated with coal bed. The lowest heat transfer coefficient of
1779.38 W/m2 K was observed in the case of a sand on base case basin.
Figure 10. Overall efficiency for 3 liters treat-

Figure 9. Effect of ceramics on coal bed. ing volume.
Figure 11. Energy utilized for 3 liters treating Figure 12. Heat transfer coefficient for 3 liters
volume. treating volume.
Table 1. Analysis of water before and after solar distillation.
Property Seawater Distillate
pH 8.2 7.6
TDS, ppm 35,430 3
TSS, ppm 30,620 8
COD, ppm 750 6
BOD, ppm 300 2
Chlorides, ppm 17,600 50
Hardness, ppm 6,620 22
649

3.1 Analysis of water
The complete analysis of both the seawater and the distillate obtained from the present study
was carried out and is shown in Table 1.
For the distillate, all the properties such as pH, TDS, TSS, COD, BOD, Chlorides and
hardness were reduced drastically, and almost distilled water was obtained.
4 CONCLUSIONS
Desalination of seawater by using the solar distillation unit is the most economical efficient
process and is also the low external energy requirement process. All the materials used for
coating were proved to be suitable materials, with improved efficiency when compared to the
base case (black paint coated solar still). The combination of base solar still basin coated
with sand and ceramics proved to be best in view of its highest efficiency (65.06%). The least
efficiency was given by black paint coated basin (42.19%). The use of paraffin wax (which
is a phase change material) and coal bed proved to be fruitful as the production of distillate
occurred even in the absence of sun until 7:30 p.m.
REFERENCES
Aburideh, H., Deliou, A., Abbad, B., Alaoui, F., Tassalit, D. & Tigrine, Z. (2012). An experimental study
of a solar still: Application on the sea water desalination of Fouka. Procedia Engineering, 33, 475–484.
Boukar, M. & Harmim, A. (2004). Parametric study of a vertical solar still under desert climatic condi-
tions. Desalination, 168, 21–28.
Dhirman, N.K. (1988). Transient analysis of a spherical solar still. Desalination, 69(1), 47–55.
Gugulothu, R., Somanchi, N.S., Devi, R.S.R. & Banoth, H. (2015). Experimental investigations on per-
formance evaluation of a single basin solar still using different energy absorbing materials. Aquatic
Procedia, 4, 1483–1491.
Krishna, P.V., Sridevi, V. & Priya, B.S.H. (2016). Comparative studies on a single slope solar distillation
unit with and without copper electroplating on aluminium basin. International Journal of Advanced
Research, 4(9), 1028–1039.
Malaeb, L., Ayoub, M.G. & Al-Hindi, M. (2014). The experimental investigation of a solar still coupled
with an evacuated tube collector. Energy Procedia, 50, 406–413.
Murugavel, K.K., Chockalingam, K.K. & Srithar, K. (2008). Progresses in improving the effectiveness
of the single basin passive solar still. Desalination, 220(1–3), 677–686.
Omara, Z.M. & Kabeel, A.E. (2014). The performance of different sand beds solar stills. International
Journal of Green Energy, 11(3), 240–254.
Panchal, H.N. (2015). Enhancement of distillate output of double basin solar still with vacuum tubes.
Journal of King Saud University-Engineering Sciences, 27(2), 170–175.
Sethi, A.K. & Dwivedi, V.K. (2013). Exergy analysis of double slope active solar still under forced cir-
culation mode. Science Direct. 51(40–42), 7394–7400.
Somanchi, N.S., Sagi, S.L.S., Kumar, T.A., Kakarlamudi, S.P.D. & Parik, A. (2015). Modelling and
analysis of single slope solar still at different water depth. Science Direct. 4, 1477–1482.
Suneesh, P.U., Paul, J., Jayaprakash, R., Kumar, S. & Denkenberger, D. (2016). Augmentation of distil-
late yield in “V”-type inclined wick solar still with cotton gauze cooling under regenerative effect.
Cogent Engineering, 3(1), 1–10.
Tanaka, H. & Nakatake, Y. (2009). One step azimuth tracking tilted-wick solar still with a vertical flat
plate reflector. Desalination, 235(1–3), 1–8.
World health organization (2017) Progress on drinking-water, sanitation and hygiene, 2017.
Yadav, Y.P. & Tiwari, G. (1987). Monthly comparative performance of solar stills of various designs.
Desalination, 67, 565–578.
650

Towards current induced magnetization reversal in magnetic nanowires
Aicy Ann Antony

P.S. Anilkumar
Department of Physics, Indian Institute of Science, Bangalore, Karnataka, India
V.O. Rejini
ABSTRACT: The effect of pulsed currents by passing through single ferromagnetic

nanowires of thickness 20 nm and width 450 nm were studied here. A pulsed current of
about 1012 A/m2 was injected at different values of the applied field. During this work,
the automation of the measurement system is carried out with a Pico pulse generator and
the output is displayed on a DPO (Digital Phosphor Oscilloscope) to get the shape of the
waveform. With this setup, transport and optimization using different pulse widths of
electrical pulses in the range 10 ns to 80 ns were carried out. The statistical analysis of these
results give information about the magnetization processes in nanowires.
1 INTRODUCTION
Spintronics or magneto-electronics is an area of active research because of the tremendous

potential both in terms of fundamental physics and technology. Here, one exploits the spin
degree of freedom of the electrons along with its charge. So we can expect new generation
of devices with completely different functionality. The advantages of these magnetic devices
would be non-volatility, increased data processing speed, decreased electric power consump-
tion and increased integration densities compared to semiconductor devices. This has led
to a revolution in magnetic data storage technology. Present day data storage heavily relies
on magnetic hard disc drives in the form of magnetic thin film media and one of the key
concerns is the magnetization reversal of the magnetic data bits by applying local magnetic
fields (1). The replacement of magnetic field assisted reversal by more sophisticated current
induced magnetic reversal mechanism is emerging as a viable and attractive option for next
generation technology.
This mechanism requires source of spin polarized electrons from a ferromagnet to switch
the magnetization. The current density required to switch the magnetization is 1012 A/m2. So
magnetic material is patterned into nanowires and hence the current requirement is reduced.
However, Joule heating will destroy these wires. Hence we need to use pulsed current of the
order of nanoseconds to prevent this damage. In this work, an experiment has been set up
to pass 10–80 ns current pulses through a magnetic nanowire (Ni81Fe19) and achieve cur-
rent induced magnetization reversal. A Pico-pulse generator will be used for this. A high
frequency oscilloscope will be used to understand the pulse shape and its distortion while
passing through the nanowire. During this work the system was automated and various
parameters were evaluated. The important instruments included in this work are the Pulse
Generator and the DPO (Digital Phosphor Oscilloscope). The programming language which
651

interfaces these Pulse Generator and DPO with the controller is LabVIEW (Laboratory
Virtual Instrument Engineering Workbench).
The Pulse Generator used here is PSPL10300B and is manufactured by Tektronix Com-
pany. It provides high amplitude positive or negative pulses and its output is designed for
50 Ω impedance. Its other features include 300 ps rise time and 750 ps fall time, adjustable
amplitude from −4 mV to 50 V in 1 db steps, adjustable duration from 1 ns to 100 ns etc:
The DPOs captures, stores, displays and analyzes in real time the three dimensions of signal
information such as Amplitude, Time and distribution of Amplitude over Time. DPO uses a
separate parallel processor. So the speed of the display is not limited. TDS794D model DPO
manufactured by Tektronix company is used here. Also its features include 2 GHz band-
width, 2 or 4 channels, advanced triggering, 1 mV to 10 V/div Sensitivity etc:
The programming language which interfaces the pulse generator and DPO with the con-
troller is the LabVIEW. It is interfaced to the instruments by using GPIB (General Purpose
Interface Bus) cable. LabVIEW is a system-design platform and developing environment for
a visual programming language from National Instruments. It is a graphical programming
language. LabVIEW programs are called virtual instruments. It is because their appearance
and operation imitate physical instruments. There is a user interface or front panel with con-
trols and indicators. A block diagram is there which contain codes to control the front panel
objects. Here we are passing 10 ns to 80 ns current pulses which are of very high frequency
from a Pulse generator to a nanowire and the magnetic reversal is studied and the result is
displayed on a digital phosphor oscilloscope.
2 METHODOLOGY
In this work the main parameters are the Pulse generator, Oscilloscope and the interfacing
language LabVIEW. The Pulse generator used here is a Pico pulse generator with 300 ps
Figure 1. Pulse generator.
Figure 2. Digital phosphor oscilloscope.
652

Figure 3. SEM image of ring device with width 450 nm and inner diameter 1.86 µm.
rise time and 750 ps fall time. This study will focus on the magnetization reversal in mag-
netic nanowires through electrical transport. The automation of the measurement system
is carried out using Pico pulse generator. With this set up the transport and optimiza-
tion using different pulse widths of electrical pulses is carried out. The statistical analy-
sis of these results are expected to give information about the magnetization processes in
nanowires.
Today, data storage heavily relies on the magnetic hard disc drives in the form of thin film.
The speed of the hard disc depends on the speed of rotation of the disc so that sensor can
read data fastly. The interchanging of north and south poles cannot be done manually. It is
done by Magnetization switching. Magnetic field is required for magnetization switching to
take place.
The replacement of magnetic field assisted reversal or switching by more sophisticated
current induced magnetic reversal is emerging as a viable and attractive option for next gen-
eration technology. This mechanism requires spin polarized electrons from a ferromagnet
to switch the magnetization. The current density required for magnetic switching is
1012 A/m2. So, for reducing the huge current, magnetic material is patterned into nanowires.
But Joule heating will destroy these wires. Hence we are using pulsed current of the order of
nanoseconds.
The above figure is the SEM image of the nanowire used to pass current pulses which is
having width 450 nm, thickness 20 nm and inner diameter 1.86 µm.
Experiments were conducted by connecting PSPL5333, which is a power divider, between

the input and output and the reflections and the current passed were studied from a high
frequency DPO (Digital Phosphor Oscilloscope). At first when trial and error has done with
different lengths of a wire it was found that as the length of the wire is reduced the attenua-
tion or reflections gets reduced and it can be understood from the voltage variations on the
DPO. At high frequency, obtained by shorter durations or pulse widths the reflection is less.
After trial and error I moved on with the sample which is Ni81Fe19. When BNC cables were
used 100% transmission was obtained. But when devices consisting of a nanoring, that is,
nanowires are connected, it is found that only 10% of the signal gets transmitted in the pulse
width range 10 ns to 80 ns.
It has been realized that this transmission is sufficient to achieve current induced
magnetization reversal in these structures. The observations done with the sample Ni81Fe19
for different pulse widths is plotted in Origin Software and is given below.
From the graph we can understand that for a pulse width of 20 ns the input signal has a
voltage of 200 mV, but after passing through the structure the signal is attenuated to 20 mV.
However the width of the pulse remain unaltered.
653

Figure 4. Input displayed on DPO by giving Figure 5. Output obtained from DPO when
10 ns pulse width through nanowire. 10 ns pulse width applied through nanowire.
Figure 6. Input displayed on DPO when 20 ns Figure 7. Output obtained from DPO when
pulse width applied through nanowire. 20 ns pulse width applied through nanowire.
654

655

4 CONCLUSION
In this work, a Pico pulse generator was studied, programmed and interfaced with the con-
troller. Also the rise time and fall time of the pulse generator was verified by passing current
pulses with 10 ns to 80 ns pulse width range. The pulse output shape and distortion were
understood from a DPO (digital phosphor oscilloscope) which is interfaced with the control-
ler. DPO was also programmed for this study. At first, when BNC cables were used for the
transmission of the current pulses, nearly 100% transmission was obtained. But when devices
consisting of a nanoring, that is, a nanowire was connected it was found that only 10% of
the signal gets transmitted in the pulse width range from 10 ns to 80 ns. So it has been real-
ized that this transmission is sufficient to achieve current induced magnetization reversal in
nanowire structures. For future scope we can say that under these conditions, we will be able
to image magnetic domain and the velocity of the magnetic domain walls at higher current
density.
REFERENCES
[1] Masamitsu Hayashi, Luc Thomas, Rai Moriya, Charles Rettner and Stuart S.P Parkin (2008),
Current-Controlled Magnetic Domain-Wall Nanowire Shift Register, Science, Vol. 320, 209–211.
[2] J.E. Wegrowe, D. Kelly, Y. Jaccard, Guittenne and Ansermet (1998), Current Induced Magnetic
reversal in magnetic nanowires, Europhysics Letters, 45, 626–631.
[3] Tao Yang, Takashi Kimura and Yoshichika Ottani (2008), Giant spin accumulation signal and pure
spin-current induced reversible magnetization switching, Nature Physics, Vol. 4, 851–854.
[4] E.A. Rando and S. Allende (2015), Magnetic reversal modes in multisegmented nanowire arrays
with long aspect ratio, Vol. 118, 013905-1 to 013905-8.
[5] A. Thiaville, Y. Nakatani, J. Miltat, Y. Suzuki (2005), Micromagnetic understanding of current-
driven domain wall motion in patterned nanowires, Europhysics Letters, Vol. 69, 990–996.
656

Synthesis and surfactant size regulation of nanoparticles

of maghemite (γ-Fe2O3)
Mushtaq Ahmad Rather

Chemical Engineering Department, National Institute of Technology, Hazratbal, Srinagar Kashmir, India
ABSTRACT: In this study, nanoparticles of maghemite (γ-Fe2O3) were synthesized

and their size regulated by utilizing different amounts of surfactant, Polyethylene Glycol
(PEG)-4000. The amount of surfactant was varied in order to analyze its effect upon the
particle size. The final synthesized particles had a mean size of 40, 23, 15, and 11 nm for
various surfactant quantities. X-ray diffraction results confirmed that the as-synthesized iron
oxide nanoparticles were γ-Fe2O3. Superconducting Quantum Interference Device (SQUID)
measurements showed the highest saturation magnetization value of 68 emu/gm at room
temperature when a surfactant quantity of 0.5 g was used under the given conditions. The
particle shape was almost a sphere, as confirmed by transmission electron microscopy.
Keywords: Nanoparticles, Iron oxide, Surfactant, Paramagnetism
1 INTRODUCTION
Iron oxides are common compounds, which are widespread in nature and can be readily syn-
thesized in the laboratory. Iron oxide nanoparticles find applications in the biomedical sec-
tor (in cellular labeling, cell separation, detoxification of biological fluids, tissue repair, drug
delivery, Magnetic Resonance Imaging (MRI), hyperthermia and magneto faction) (Gupta &
Gupta, 2005), electrode materials (Kijima et al., 2011), fabrication of pigments, sorbents, gas
sensors (Afkhami & Moosavi, 2010), ferro fluids and wastewater purification (Shen, 2009a).
Eight iron oxides are known. Among these, hematite (α-Fe2O3), magnetite (Fe3O4) and
maghemite (γ-Fe2O3) are very promising and popular candidates due to their polymor phism
involving temperature-induced phase transition. These three iron oxides have unique prop-
erties (such as biochemical, magnetic, and catalytic), which make them suitable for specific
technical and biomedical applications (Cornell & Schwertmann, 2003). Maghemite is fer-
rimagnetic at room temperature but its nanoparticles smaller than 10 nm are superparamag-
netic. Maghemite is unstable at high temperatures and loses it susceptibility with time (Ray
et al., 2008; Neuberger et al., 2005). The structure of γ-Fe2O3 is cubic. Oxygen anions give rise
to a cubic close-packed array while ferric ions are distributed over tetrahedral sites (eight Fe
ions per unit cell) and octahedral sites (the remaining Fe ions and vacancies). Therefore, the
maghemite can be considered as being fully oxidized magnetite, and it is an n-type semicon-
ductor with a band gap of 2.0 eV (Wu et al., 2015).
Particle agglomeration forms large clusters, resulting in an overall increase in particle size
(Hamley, 2003). When two large-particle clusters approach one another, each of them comes
under the influence of the magnetic field of its neighbor. Besides the arousal of attractive
forces between the particles, each particle is within the magnetic field of the neighbor and
becomes further magnetized (Tepper et al., 2003). The adherence of remnant magnetic particles
causes a mutual magnetization, resulting in increased aggregation properties. Since particles
are attracted magnetically in addition to the usual flocculation due to Van der Waals forces,
surface modification is often indispensable. For effective stabilization of iron oxide nanopar-
ticles, often a very high requirement of density for coating is desirable. Some stabilizer such as
657

a surfactant or a polymer is usually added at the time of preparation to prevent aggregation
of the nanoscale particulates. Most of these polymers adhere to surfaces in a substrate-specific
manner (Mendenhall et al., 1996). Nanoparticle coatings may comprise of several materials,
including both inorganic and polymeric materials. Polymeric coating materials can be classified
into synthetic and natural types. Polymers based on, for instance, poly (ethylene-co-vinyl ace-
tate), Poly (Vinyl Pyrrolidone) (PVP), Poly (Lactic Co-Glycolic Acid) (PLGA), Poly (Ethylene
Glycol) (PEG), and Poly (Vinyl Alcohol) (PVA), are typical examples of synthetic polymeric
systems (Miller et al., 1983; Ruiz & Benoit, 1991). Natural polymer systems include the use of
substances such as gelatin, dextran, chitosan, and pullulan (Li et al., 1997; Massa et al., 2000).
As we know the addition of surfactant time and the amount that would regulate the growth
of particles, in the present study PEG-4000 was used as a surfactant to stabilize the particles
in aqueous medium. Different amounts of this surfactant were added to obtain Nanoparti-
cles (NPs) of various sizes and to analyze its effect upon the particle size.
2 EXPERIMENTAL
Synthesis of nanoparticles was carried out using a coprecipitation process based on work
by Yunabi et al. (2008), Shen et al. (2009b), and Maity and Agrawal (2007). The modified
procedure adopted is briefly stated as follows.
A solution of 50 ml of ferrous chloride tetrahydrate was added to 100 ml of ferric chloride
hexahydrate solution under constant agitation. The contents were heated to 50°C in a Borosil
glass round-bottom reaction vessel. The 500 ml capacity reactor was specially designed for
the purpose. Ammonium hydroxide solution was added drop-wise at the specified tempera-
ture. The initially black-brown colored solution formed changed into a completely black
mixture after about half an hour at a pH of 9. The magnetic particles so formed were kept
dispersed by an ultra-sonicator for about a further 15 minutes and then collected by settling
under a ferro magnet. The particles were then transferred into a separate beaker and washed
with water until the pH reached 7. Then the particles were washed twice with ethanol. The
collected nanoparticles were dried in a vacuum oven at 60°C for about six hours.
PEG-4000 was used as surfactant to stabilize the particles in aqueous medium. In order to
observe the effect of adding different amounts of PEG-4000 upon the final particle size, dif-
ferent 0.5 g, 1.0 g, 1.5 g, and 2.0 g amounts of it were used. It is pertinent to mention the fact
that an increase of surfactant time and amount would regulate the growth of the particles.
Four of the γ-maghemite NPs (S1, S2, S3, and S4) were obtained by adding PEG-4000 sur-
factant amounts as 0.5 g, 1 g, 1.5 g, and 2.0 g respectively. Crystallite size, magnetic properties
and surface area of these NPs were determined.
The size of the particles and their shape were observed by Transmission Electron Micros-
copy (TEM). The type of crystal was verified by X-Ray Diffraction (XRD) study after com-
parison of the same with the JCPDS file. Standard patterns for bulk magnetite, maghemite,
and hematite are respectively given by JCPDS file numbers. 19-0629, 39-1346 and 83-0664.
The crystallite size was calculated by using the Scherrer equation (Paterson, 1939). The
magnetic properties of the nanoparticles were measured by Superconducting Quantum
Interference Device (SQUID) measurements (Tepper et al., 2003). A Malvern Dynamic Light
Scattering (DLS) nanosizer was used to analyze the size of the particles in solution.
3.1 Transmission Electron Microscopic (TEM) results

Interpretation of TEM pictures reveals the average size of particles to be 40, 23, 15, and
11 nm respectively for S1, S2, S3, and S4. Hence, as surfactant quantity increases from 0.5 g
to 2.0 g, the particle size decreases from 40 to 11 nm. The photographs are shown in Figure 1
below.
658

Figure 1. TEM images: (a) S1; (b) S2; (c) S3; (d) S4.
Figure 2. DLS images: (a) S1; (b) S2; (c) S3; (d) S4.
3.2 Dynamic Light Scattering (DLS) analysis

The DLS images reveal the average size of particles to be 90, 60, 38, and 24 nm respectively
for S1, S2, S3, and S4. Magnetic particles have a very high tendency of agglomeration due
to the presence of additional magnetic interactions, in addition to the usual Van der Waals
forces of attraction. Their size may not exactly correspond to that indicated by the TEM due
to this agglomeration of NPs. The images are shown in the following Figure 2.
3.3 X-Ray Diffraction (XRD) analysis

In all the cases of S1, S2, S3, and S4, the bulk phase identified is maghemite (i.e. γ-Fe2O3).
The majority of peaks in the XRD pictures and the peak of 100% intensity correspond to the
γ-maghemite. Some of the peaks also are nearer to hematite, so it may also be present as an
oxidation product together with γ-maghemite. The XRD pictures are shown in Figure 3, giv-
ing 2θ versus intensity data for the above γ-maghemite NPs S1, S2, S3, and S4. The crystallite
size computed by the Scherrer equation was 35, 20, 19, and 10 nm respectively.
Peak-wise analysis for the identification of type of crystallite in S1, S2, S3, and S4, showed
the predominant presence of maghemite (γ-Fe2O3) in each case. The above analysis clearly
showed the presence of the γ-maghemite phase of iron oxide NPs. The presence of other
phases was not ruled out as the 2θ versus intensity data for different iron oxide phases are
very close, and particularly those for magnetite and maghemite.
3.4 SQUID measurements

SQUID measurements of samples S1, S2, S3, and S4 show respectively saturation magnetiza-
tion values of 68, 56.5, 54.87, and 30.18 emu/gm at room temperature. This indicates that as
659

Figure 3. XRD: (a) S1; (b) S2; (c) S3; (d) S4.
Figure 4. Hysteresis curves: (a) S1; (b) S2; (c) S3; (d) S4.
the PEG-4000 surfactant amount increases, the size of the NPs and saturation magnetization
value decrease. The saturation magnetization value of bulk maghemite is about 73 emu/gm
(Maity & Agrawal, 2007), and the above values compare well with it. Figure 4 below shows
the hysteresis curves of S1, S2, S3, and S4. All the magnetic hysteresis loops have passed the
grid origin, which indicates good super paramagnetism.
3.5 Brunauer, Emmett and Teller (BET) surface area studies

The values for different NPs of γ- maghemite were 54.31, 61.8, 66.4, and 69.2 m2/gm respec-
tively for S1, S2, S3, and S4.
4 CONCLUSION
In the given technique, four sizes of γ-maghemite NPs, viz. 40, 23, 15, and 11 nm, were
obtained by using PEG-4000 surfactant amounts of 0.5 g, 1 g, 1.5 g, and 2.0 g respectively
under the given conditions. The crystallite size, magnetic properties and surface area of these
NPs were determined. PEG-4000 surfactant was successful in avoiding the huge agglomera-
tion of NPs in solution. SQUID measurements showed the highest saturation magnetiza-
tion value to be 68 emu/gm at room temperature when the surfactant quantity was 0.5 g. As
660

surfactant amount was increased, the size of the NPs and the saturation magnetization value
decreased. The synthesized NPs will be useful in various practical applications.
ACKNOWLEDGMENT
The author acknowledges the support received from National Institute of Technology, Srina-
gar, Kashmir for facilitating the work.
REFERENCES
Afkhami, A. & Moosavi, R. (2010). Adsorptive removal of Congo red, a carcinogenic textile dye, from
aqueous solutions by maghemite nanoparticles. J. Hazard. Mater., 174(1–3), 398–403.
Cornell, R.M. & Schwertmann, U. (2003). The iron oxides: Structure, properties, reactions, occurrences
and uses (2nd ed.). Weinheim: Wiley.
Gupta, A.K. & Gupta, M. (2005). Synthesis and surface engineering of iron oxide nanoparticles for
biomedical applications. Biomaterials, 26(18), 3995–4021.
Hamley, I.W. (2003). Nanotechnology with soft materials. Angew Chem. Int. Ed. Engl, 42(15),
1692–1712.
Kijima, N., Yoshinaga, M., Awaka, J. & Akimoto, J. (2011). Microwave synthesis, characterization, and
electrochemical properties of α-Fe2O3 nanoparticles. Solid State Ionics, 192(1), 293–297.
Li, J.K., Wang, N. & Wu, X.S. (1997). A novel biodegradable system based on gelatin nanoparticles and
poly (lactic-co-glycolic acid) microspheres for protein and peptide drug delivery. Journal of Pharma-
ceutical Sciences, 86(8), 891–895.
Massia, S.P., Stark, J. & Letbetter, D.S. (2000). Surface-immobilized dextran limits cell adhesion and
spreading. Biomaterials, 21(22), 2253–2261.
Maity, D. & Agrawal, D. (2007). Synthesis of iron oxide nanoparticles under oxidizing environment and
their stabilization in aqueous and non-aqueous media. Journal of Magnetism and Magnetic materials,
308(1), 46–55.
Mendenhall, G.D., Geng, Y. & Hwang, J. (1996). Mendenhall, G.D., Geng, Y., & Hwang, J. (1996).
Optimization of long-term stability of magnetic fluids from magnetite and synthetic polyelectrolytes.
Journal of Colloid and Interface Science, 184(2), 519–526.
Miller, E.S., Peppas, N.A. & Winslow, D.N. (1983). Morphological changes of ethylene/vinyl acetate-
based controlled delivery systems during release of water-soluble solutes. Journal of Membrane Sci-
ence, 14(1), 79–92.
Neuberger, T., Schöpf, B., Hofmann, H., Hofmann, M. & Von Rechenberg, B. (2005). Superparamag-
netic nanoparticles for biomedical applications: Possibilities and limitations of a new drug delivery
system. Journal of Magnetism and Magnetic materials, 293(1), 483–496.
Paterson, A.L. (1939). The Scherrer formula for X-ray particle size determination. Physical Review,
56(10), 978–982.
Ray, I., Chakraborty, S., Chowdhury, A., Majumdar, S., Prakash, A., Pyare, R. & Sen, A. (2008). Room
temperature synthesis of γ-Fe2O3 by sonochemical route and its response towards butane. Sens Actua-
tors B Chem, 130(2), 882–888.
Ruiz, J.M. & Benoit, J.P. (1991). In vivo peptide release from poly (DL-lactic acid-co-glycolic acid)
copolymer 5050 microspheres. J Control Release, 16(1–2), 177–185.
Shen, Y.F., Tang, J., Nie, Z.H., Wang, Y.D., Ren, Y. & Zuo, L. (2009a). Preparation and application of
magnetic Fe3O4 nanoparticles for wastewater purification. Separation and Purification Technology,
68(3), 312–319.
Shen, Y.F., Tang, J., Nie, Z.H., Wang, Y.D., Ren, Y. & Zuo, L. (2009b). Tailoring size and structural
distortion of Fe3O4 nanoparticles for the purification of contaminated water. Bioresource Technol-
ogy, 100(18), 4139–4146.
Tepper, T., Ilievski, F., Ross, C.A., Zaman, T.R., Ram, R.J., Sung, S.Y. & Stadler, B.J.H. (2003). Faraday
activity in flexible maghemite/polymer matrix composites. J Appl Phys, 93(10), 6948–6950.
Wu, W., Wu, Z., Yu, T., Jiang, C. & Kim, W.S. (2015). Recent progress on magnetic iron oxide nanopar-
ticles: Synthesis, surface functional strategies and biomedical applications. Sci Technol Adv Mater,
16(2), 023501.
Yunabi, Z., Zumin, Q. & Huang, J. (2008). Preparation and analysis of Fe3O4 magnetic nanoparticles
used as targeted drug carriers. Chinese Journal of Chemical Engineering, 16(3), 451–455.
661

Hydrothermal carbonization: A promising transformation process

of biomass into various product materials
Mushtaq Ahmad Rather

Chemical Engineering Department, National Institute of Technology, Hazratbal, Srinagar, Kashmir,
India
ABSTRACT: Hydrothermal Carbonization (HTC) conversion involves the reaction of bio-

mass or other waste organic materials in water at high temperature and pressure, near the ther-
modynamic critical point of water (Tc = 373.95°C, Pc = 22.064 MPa) in a specially designed
high-pressure autoclave reactor. HTC conversion leads to the formation of solid Hydrochar
(HC) as a predominant product. This product is a versatile material with potential uses such
as a fuel, adsorbent, or anode in Li-ion batteries. The present paper summarizes some of the
important aspects of this recently emerging field in manufacturing and technology.
Keywords: Hydrothermal Carbonization, Hydrochar
1 INTRODUCTION
Hydrothermal carbonization (HTC) is a 100-year-old technique in the synthesis of materi-

als, with increasing interest originating from the formation of charcoal. The Bergius process,
firstly developed by German chemist Friedrich Bergius in 1913, is a method of production
of liquid hydrocarbons for use as synthetic fuel by hydrogenation of high-volatile bitu-
minous coal at high temperature and pressure (Hu et al., 2010). He was also awarded the
Nobel Prize in Chemistry in 1931 for his development of high-pressure chemistry (Bergius,
1996). The pyrolysis in the presence of subcritical liquid water is called hydrothermal car-
bonization (Tekin et al., 2014). The process generates both solid and liquid products. The
solid fuel generated in HTC conversion is referred to as Hydrochar (HC) (biochar). HTC
of lignocellulosic biomass has received extensive research over the last two decades for
both the production of solid and liquid fuels (subcritical conditions), and for gasification
(supercritical conditions).
Hot compressed liquid water near its thermodynamic critical point (Tc = 373.95°C,
Pc = 22.064 MPa) behaves very differently from liquid water at room temperature. As water
is heated along its vapor–liquid saturation curve, its dielectric constant decreases due to
the hydrogen bonds between water molecules being fewer and less persistent. The reduced
dielectric constant enables hot compressed water to solvate small organic molecules, allow-
ing organic reactions to occur in a single fluid phase. Additionally, the ion product of water
increases with temperature up to about 280°C, but then decreases as the critical point is
approached. This higher ion product leads to higher natural levels of hydronium ions in
hot compressed water, which can accelerate the rates of acid-catalyzed hydrolytic decom-
position reactions (Yeh et al., 2013). Hydrothermal carbonization processing wet biomass
can produce a hydrochar that retains a large proportion of the chemical energy and lipids
in the original biomass. The hydrothermal carbonization environment promotes the hydro-
lytic cleavage of ester linkages in lipids, peptide linkages in proteins, and glycosidic ether
linkages in carbohydrates. These cleavage reactions can be accelerated by catalysts (Yeh
et al., 2013).
663

Figure 1. Schematic diagram of hydrothermal carbonization reactor setup.
2 EXPERIMENTAL
The HTC of biomass material to fine-powdered form is carried out in an autoclave reac-
tor, usually made up of alloy steel. Controlled heating by a temperature controller is car-
ried out by means of a heater of suitable power rating. The reactor is properly sealed and
head-tightened for leak-proof operation. Under inert atmosphere, heating is started, and
the autogenic pressure builds up. At the desired temperature, the carbonization reaction is
allowed to proceed for the holding time for a suitable duration. The heating is then stopped
and the reactor allowed to cool to room temperature. The carbonized solids are then sepa-
rated from the liquid phase by means of vacuum filtration. The solid-product hydrochar is
dried in an oven to remove any residual moisture. Dried HC is weighed and processed for
further characterization. The schematic diagram of the experimental setup for HTC conver-
sion is illustrated in Figure 1.
3 PROGRESS IN HTC PROCESS
Ozcimen et al. (2008) performed carbonization experiments on grape seed and chestnut shell
samples, having the average particle size of 0.657 mm and 0.377 mm respectively, to deter-
mine the effect of temperature, sweep gas flow rate and heating rate on the biochar yield. It
was found that the temperature had the dominant effect on the biochar yields, as compared
to the effects of nitrogen gas flow rate and heating rate.
Sevilla and Fuertes (2009) produced highly functionalized carbonaceous materials by
means of the HTC of cellulose at temperatures in the range of 220–250°C. They observed
that the materials so formed were composed of agglomerates of carbonaceous microspheres
(size 2–5 µm) as evidenced by scanning electron microscopy.
Funke and Ziegler (2010) elaborated the reaction mechanisms of hydrolysis, dehydration,
decarboxylation, aromatization, and condensation polymerization during HTC. The mecha-
nisms were important in studying the role of different operational parameters qualitatively
for cellulose, peatbog and wood. The results were used to derive fundamental process design
improvements for HTC.
Anastasakis and Ross (2011) subjected the brown macro-alga Laminaria saccharina
to hydrothermal carbonization conversion for the generation of solid and liquid biofuels.
Experiments were performed in a batch bomb-type stainless steel reactor (75 ml). The heat-
ing rate of the reactor was 25 K/min. The reactor was charged with the appropriate amounts
of seaweed biomass and water. In the catalytic runs, an appropriate amount of KOH was
added to the reactants. The influence of reactor loading, residence time, temperature and
catalyst (KOH) loading were assessed. The experimental conditions were found to have a
664

profound effect on the resulting char composition. The carbon content in the chars ranged
from 14.9–52.6 wt.%.
Hoekman et al. (2011) in their experimental effort used a 2 L pressure vessel to apply the
HTC process to a mixed wood feedstock. The effects of the reaction conditions on product
compositions and their yields were examined by varying temperature over the range of 215–
295°C and the reaction hold time over the range of 5–60 min. With increasing temperature
and time, the amounts of gaseous products and produced water increased, while the amount
of HTC char decreased.
Liu et al. (2012) studied HTC to upgrade waste biomass (coconut fiber and eucalyptus
leaves) at temperatures ranging from 150–375°C and a residence time of 30 min. The kinetics
of HTC was investigated and the fuel properties of the produced hydrochars were evaluated.
Roman et al. (2012) studied the HTC processes under different conditions using two dif-
ferent biomass materials: walnut shell and sunflower stem. Walnut shell and sunflower stem
were subjected to HTC in order to increase their energy content, to provide a greater amount
of energy per unit of mass. The hydrothermal carbonization processes were carried out using
5 g of raw material dispersed in deionized water (100–150 mL) in a stainless steel autoclave,
which was heated up in an electric furnace at selected temperatures (190–230°C), during time
intervals of 20–45 h. It was found that the temperature and water/biomass ratio were more
important variables than the residence time for HTC.
Reza et al. (2013) used HTC as a pretreatment process to convert loblolly pine as feedstock
to homogeneous energy-dense solid fuels. Experiments were conducted in a two-chamber
reactor, maintaining isothermal conditions for 15 s to 30 min reaction times. Loblolly pine
was treated at 200, 230, and 260°C. In the first few minutes of reaction, the solid-product
mass yield decreased rapidly while the calorific value increased rapidly.
Wiedner et al. (2013) focused on chemical modification of wheat straw, poplar wood
and olive residues through HTC at different temperatures (180°C, 210°C, and 230°C).
Besides general properties such as pH, Electrical Conductivity (EC), ash content, elemen-
tal composition and yield, they evaluated bulk chemical composition and the contribution
of specific compounds (lignin and black carbon). Hydrochar yields and carbon recovery
decreased with increasing temperature to about 50% and 75%, respectively for all feed-
stocks at 230°C.
Eibisch et al. (2013) carbonized biomass via HTC to yield a Carbon (C) rich hydrochar.
They investigated whether easily mineralizable organic components adsorbed on the hydro-
char surface influenced the degradability of the hydrochars and so could be removed by
repetitive washing.
Reza et al. (2014) in their study on HTC of cellulose, wheat straw, and poplar, evaluated
the effects of reaction temperature and reaction time on both solid hydrochar and process
liquid. The objective was to design a high pressure, high temperature slurry sampling system
in an 18.6 L reactor. Several different reaction conditions by changing biomass feedstock,
reaction time, and reaction temperature was studied. Hydrochar composition, as well as
chemical components in HTC, varied with temperature and time.
Using HTC conversion, Zhao et al. (2014) produced nitrogen- and chlorine-free solid bio-
fuel from high moisture and nitrogen content bio-wastes, such as Municipal Solid Waste
(MSW), mycelia waste, sewage sludge and paper sludge. The work focused on energy recy-
cling, and on optimizing the operating parameters and evaluating the energy efficiency of
this fuel production process. The effect of the temperature and holding time on the biofuel
recovering ratio, calorific value and energy recovery rate were investigated.
Guiotoku et al. (2014), in their work on hydrothermal carbonization, carbonized cellulose-
based renewable raw materials in a microwave oven at 200°C for 60, 120, and 240 min. The
values of fixed carbon were between 38 and 52%.
Danso-Boateng et al. (2015a) carried out HTC of primary sewage sludge using a batch
reactor to convert wet biomass (sewage sludge) to lignite-like renewable solid fuel of high
calorific value. The effect of temperature and reaction time were investigated on the char-
acteristics of solid hydrochar, liquid and gas products, and the conditions leading to opti-
mal hydrochar characteristics. The amount of carbon retained in hydrochars decreased as
665

temperature and time increased, with carbon retentions of 64–77% at 140 and 160°C, and
50–62% at 180 and 200°C.
Danso-Boateng et al. (2015b), in their work on sewage sludge, carried out the mass and
energy balances of semi-continuous HTC of fecal waste at 200°C and at a reaction time of
30 min.
Lin et al. (2015) used different temperatures in the range of 180–300°C to evaluate the
effect of HTC temperature on hydrochar fuel characteristics and their thermal behavior. The
hydrochar produced at 210°C had the maximum heating value (9,763 kJ/kg) with the highest
energetic recovery efficiency (90.12%). Therefore, 210°C was the optimum temperature for
HTC of paper sludge.
Yin et al. (2015) conducted batch hydrothermal carbonization tests for hydrothermal car-
bonization decomposition of sewage sludge using a tubular reactor (316 L stainless steel)
with 5 mm internal diameter and 8.2 ml volume.
Benavente (2015) focused in the application of the HTC technology as a possible waste
management treatment of moist agro-industrial waste. Through this technique, Olive Mill,
Canned Artichoke and Orange Wastes (OMW, CAW and OW, respectively) were carbon-
ized in a lab-scale high-pressure reactor at different temperatures (200–250°C) and durations
(2, 4, 8, and 24 h) in order to obtain useful bioenergy feedstocks.
Pruksakit and Patumsawad (2016) investigated the HTC of sugar cane at different temper-
atures, (189, 220, and 250°C), and for different retention times (0.1, 1, and 2 h), to study the
effect of operating conditions upon product yield and some other parameters. With increas-
ing temperature, hydrochar yield, hydrogen and oxygen decreased. However, the carbon con-
tent of hydrochar increased.
Mau et al. (2016) studied the HTC of poultry litter under a range of process parameters
(temperature, reaction time, and solids concentration). Results showed the production of
hydrochar with caloric value of 24.4 MJ/kg, similar to sub-bituminous coal. Temperature
had the most significant effect on processes and product formation. Solids concentration was
not a significant factor once dilution effects were considered.
Wikberg et al. (2016) studied the progress of the conversion, the yield, the structure and
the morphology of the produced carbonaceous materials as a function of time. Carbona-
ceous particles of different shape and size were produced with yields between 23% and 73%
after 4 h, with the yields being higher for lignin than for carbohydrates. According to the
results, potential pulp mill streams represent lignocellulosic resources for the generation of
carbonaceous materials.
Lin et al. (2016) studied the resource utilization of MSW at several temperatures (210,
230, 250, and 280°C) and residence times (30, 60, and 90 min) to investigate the effects on
the characteristics of HTC solid fuel from MSW. The results of thermogravimetric analysis
illustrated that HTC did remarkably influence the thermal behavior and kinetics of MSW. In
most cases, the HTC temperature had a more obvious effect than residence time.
Nizamuddin (2016) investigated the possible optimum conditions for maximum yield of
hydrochar through HTC of palm shell. The hydrochar and the palm shell were characterized,
and the chemical, dielectric, and structural properties of optimized hydrochar were exam-
ined. The effects of the reaction temperature, reaction time and biomass-to-water ratio were
analyzed and optimized using the central composite design of response surface methodology.
The optimized conditions for hydrochar production were found to be 180°C, 30 min, and
1.60 wt.%, temperature, time, and biomass-to-water ratio, respectively.
4 CONCLUSION
HTC in coming days is to play a great role in the transformation of waste biomass into ver-
satile products that may for instance be in the form of fuels, adsorbents, anode in Li-ion bat-
teries, and soil enrichment agents. We need to carry out an exhaustive study of biomass and
its types that are available in India so that if the need arises, the hydrothermal carbonization
may be used for transformation into various useful products.
666

ACKNOWLEDGMENT
The author acknowledges the support received from National Institute of Technology, Srinagar,
Kashmir for facilitating the work.
REFERENCES
Anastasakis, K., & Ross, A.B. (2011). Hydrothermal liquefaction of the brown macro-alga Laminaria
saccharina: effect of reaction conditions on product distribution and composition. Bioresource tech-
nology, 102(7), 4876–4883.
Benavente, V., Calabuig, E., & Fullana, A. (2015). Upgrading of moist agro-industrial wastes by hydro-
thermal carbonization. Journal of Analytical and Applied Pyrolysis, 113, 89–98.
Bergius, F. (1996). Chemical reactions under high pressure. Nobel Lectures, Chemistry 1922–1941,
244–276.
Danso-Boateng, E., Shama, G., Wheatley, A.D., Martin, S.J., & Holdich, R.G. (2015). Hydrothermal
carbonisation of sewage sludge: Effect of process conditions on product characteristics and methane
production. Bioresource technology, 177, 318–327.
Danso-Boateng, E., Holdich, R.G., Martin, S.J., Shama, G., & Wheatley, A.D. (2015). Process ener-
getics for the hydrothermal carbonisation of human faecal wastes. Energy Conversion and Manage-
ment, 105, 1115–1124.
Eibisch, N., Helfrich, M., Don, A., Mikutta, R., Kruse, A., Ellerbrock, R., & Flessa, H. (2013). Pro
perties and degradability of hydrothermal carbonization products. Journal of environmental qual-
ity, 42(5), 1565–1573.
Funke, A., & Ziegler, F. (2010). Hydrothermal carbonization of biomass: a summary and discussion of
chemical mechanisms for process engineering. Biofuels, Bioproducts and Biorefining, 4(2), 160–177.
Guiotoku, M., Rambo, C.R., & Hotza, D. (2014). Charcoal produced from cellulosic raw materi-
als by microwave-assisted hydrothermal carbonization. Journal of Thermal Analysis and Calori
metry, 117(1), 269–275.
Hoekman, S.K., Broch, A., & Robbins, C. (2011). Hydrothermal carbonization (HTC) of lignocellulosic
biomass. Energy & Fuels, 25(4), 1802–1810.
Hu, B.B., Wang, K., Wu, L., Yu, S.H., Antonietti, M. & Titirici, M.M. (2010). Engineering carbon mate-
rials from the hydrothermal carbonization process of biomass. Advanced Materials, 22(7), 813–828.
Lin, Y., Ma, X., Peng, X., Hu, S., Yu, Z., & Fang, S. (2015). Effect of hydrothermal carbonization tem-
perature on combustion behavior of hydrochar fuel from paper sludge. Applied Thermal Engineer-
ing, 91, 574–582.
Lin, Y., Ma, X., Peng, X., Yu, Z., Fang, S., Lin, Y., & Fan, Y. (2016). Combustion, pyrolysis and char
CO2-gasification characteristics of hydrothermal carbonization solid fuel from municipal solid
wastes. Fuel, 181, 905–915.
Liu, Z., & Balasubramanian, R. (2012). Hydrothermal carbonization of waste biomass for energy gen-
eration. Procedia Environmental Sciences, 16, 159–166.
Mau, V., Quance, J., Posmanik, R., & Gross, A. (2016). Phases’ characteristics of poultry litter hydro-
thermal carbonization under a range of process parameters. Bioresource technology, 219, 632–642.
Nizamuddin, S., Mubarak, N.M., Tiripathi, M., Jayakumar, N.S., Sahu, J.N., & Ganesan, P. (2016).
Chemical, dielectric and structural characterization of optimized hydrochar produced from hydro-
thermal carbonization of palm shell. Fuel, 163, 88–97.
Özçimen, D., & Ersoy-Meriçboyu, A. (2008). A study on the carbonization of grapeseed and chestnut
shell. Fuel Processing Technology, 89(11), 1041–1046.
Pruksakit, W., & Patumsawad, S. (2016). Hydrothermal Carbonization (HTC) of Sugarcane Stranded:
Effect of Operation Condition to Hydrochar Production. Energy Procedia, 100, 223–226.
Reza, M.T., Wirth, B., Lüder, U., & Werner, M. (2014). Behavior of selected hydrolyzed and dehydrated
products during hydrothermal carbonization of biomass. Bioresource technology, 169, 352–361.
Reza, M.T., Yan, W., Uddin, M.H., Lynam, J.G., Hoekman, S.K., Coronella, C.J., & Vásquez, V.R.
(2013). Reaction kinetics of hydrothermal carbonization of loblolly pine. Bioresource technol-
ogy, 139, 161–169.
Román, S., Nabais, J.M.V., Laginhas, C., Ledesma, B., & González, J.F. (2012). Hydrothermal car-
bonization as an effective way of densifying the energy content of biomass. Fuel Processing Technol-
ogy, 103, 78–83.
Sevilla, M., & Fuertes, A.B. (2009). The production of carbon materials by hydrothermal carbonization
of cellulose. Carbon, 47(9), 2281–2289.
667

Tekin, K., Karagöz, S. & Bektaş, S. (2014). A review of hydrothermal biomass processing. Renewable
and Sustainable Energy Reviews, 40, 673–687.
Wiedner, K., Naisse, C., Rumpel, C., Pozzi, A., Wieczorek, P., & Glaser, B. (2013). Chemical modifica-
tion of biomass residues during hydrothermal carbonization–What makes the difference, temperature
or feedstock? Organic Geochemistry, 54, 91–100.
Wikberg, H., Ohra-aho, T., Honkanen, M., Kanerva, H., Harlin, A., Vippola, M., & Laine, C. (2016).
Hydrothermal carbonization of pulp mill streams. Bioresource technology, 212, 236–244.
Yeh, T.M., Dickinson, J.G., Franck, A., Linic, S., Thompson Jr, J.G. & Savage, P.E. (2013). Hydrother-
mal catalytic production of fuels and chemicals from aquatic biomass. Journal of Chemical Technol-
ogy and Biotechnology, 88(1), 13–24.
Yin, F., Chen, H., Xu, G., Wang, G., & Xu, Y. (2015). A detailed kinetic model for the hydrothermal
decomposition process of sewage sludge. Bioresource technology, 198, 351–357.
Zhao, P., Shen, Y., Ge, S., & Yoshikawa, K. (2014). Energy recycling from sewage sludge by producing
solid biofuel with hydrothermal carbonization. Energy conversion and management, 78, 815–821.
668

Electronics, Signal Processing and Communication
Engineering (E-SPACE)

Denoising of musical signals using wavelets specific for

musical instruments
P.V. Sreelakshmi & A. Gayathri

Department of Electronics and Communication Engineering, Government Engineering College,
Thrissur, Kerala, India
M.S. Sinith
ABSTRACT: For musical signals, a waveform of a single note has a repeating element, as
it contains fundamental frequency and its harmonics. A wavelet designed specifically for a
musical instrument by taking this waveform as the scaling function can be used to analyze
these musical signals. Since the waveform of a single note which is used as a scaling func-
tion does not satisfy orthogonality property, they can be designed as biorthogonal wavelets.
In this paper, the filter bank coefficients corresponding to this wavelet are derived from the
available analysis low-pass coefficients using the properties satisfied by biorthogonal wave-
let. The musical signals can be decomposed and reconstructed using this set of filter bank
coefficients. The coefficients thus obtained are modified using lifting technology for better
performance. The lifting scheme is an approach to construct so-called second generation
wavelets, which are not necessarily transalates and dilations of one function. Signal being
corrupted with noise is found to be a major problem in signal processing. The musical signals
are denoised using classical wavelets and two sets of filter bank coefficients obtained using
the two methods. The denoising is performed by adopting a proper thresholding method.
For the performance comparison and measurement of quality of denoising, the Signal to
Noise Ratio (SNR) is calculated between original musical signal and the denoised signal. It is
found that coefficients give better performance once modified using lifting technology.
Keywords: Wavelets, Biorthogonal Wavelets, Filter Bank, Thresholding, Denoising
1 INTRODUCTION
Transmitted signals are mostly corrupted by noise. Once corrupted by noise, a signal loses
its pure signal characteristics. Recovering these characteristics from the corrupted signal is a
major challenge in the signal processing area. The wavelet transform technique is a widely used
method to denoise signal since it gives better results. Wavelet transform replaces Fourier trans-
form in analyzing non-stationary signals. It analyzes a signal by truncating the signal using a
window which has variable time frequency resolution called a wavelet. Daubechies introduced
the wavelet transform as a tool that cuts up data or functions or operators into different fre-
quency components, and then studied each component with a resolution matched to its scale
[1]. Eventhough wavelet analysis replaces Fourier analysis, it is a natural extension of it. Wave-
lets have been called a mathematical microscope; compressing wavelets increases the magnifica-
tion of this microscope, enabling us to take a closer look at small details in the signal [2]. The
theory of wavelet analysis and design of the filter bank coefficients are given in [3]. The signals
produced by musical instruments are found to be non stationary signals where small duration
signals or small band-width musical pieces are placed at an effective temporal position to give
special effects. Wavelet transform serves as a good technique to analyse those signals. In the
671

existing wavelets, Coiflet5 is found to be the most suitable wavelet for analyzing musical signals.
For the best analysis of a signal of interest, it is desirable to design a wavelet that matches the
signal. Algorithms for designing wavelets that match the signal of interest are given in [4]. The
scaling function and wavelet function of the wavelets present in the different signals are differ-
ent. In the case of musical signals, the waveform of a single note played by a musical instrument
has a repeating element. From this repeating element, the scaling and wavelet functions for that
particular instrument can be found. This repeating element is taken as the scaling function and
hence it should satisfy the necessary and sufficient conditions for a scaling function. There are
different algorithms to derive the the filter coefficients corresponding to a scaling function.
They are Adaptive filter algorithms like LMS [5], NLMS [6] or Recursive Least Square (RLS).
Inorder to enhance the accuracy of results, there are several other variants of this algorithms
[7]–[11]. Sinith et al. used the LMS algorithm to find out the filter coefficients. But the algo-
rithm is very sensitive to step size [12]. NLMS and RLS algorithm are not sensitive to step size
and worked well for standard wavelets. Eventhough adaptive filter algorithms work well for
known wavelet, they fail to give satisfactory results for musical signals whose equations are
unknown. This leads to adopting a method known as Particle Swarm Optimisation technique
[13]–[14]. Hence the filter coefficients corresponding to the scaling function can be found accu-
rately using PSO (Particle swarm optimization) algorithm. The algorithm is modified For bet-
ter results. Thus using modified PSO algorithm, optimum values for filter coefficients, h(n) are
obtained that corresponds to a scaling function and this results in a new wavelet for analyzing
musical signals known as SSM (Sinith-Shikha-Murthy)Wavelet [15].
Once the filter bank coefficient, h(n), is obtained, the scaling function, f(t) and the wavelet
function, ψ(t) can be obtained using the relation given below.
∞
φ (t ) = ∑ h( n)φ (2t − n)
n = −∞
(1)
∞
ψ (t ) = ∑ h( n)φ (2t − n)
n = −∞
(2)
where
g ( n ) = ( −1)n h(1 − n ) (3)
The method of generation of f(t) is as shown in Figure 1. The input is an impulse function
and after a few iterations the output obtained will be the scaling function, f(t), as per the
Equations given above.
In the case of classical wavelets, the analysis and synthesis filter bank coefficients are the
same, since at a given scale the shifted versions of the scaling function and wavelet function
are orthogonal to each other. The conditions are true for the case of standard wavelets like
Daubechies and Morlet. However in the case of biorthogonal wavelets, analysis and synthesis
Figure 1. Method for finding f(t) iteratively.
672

Figure 2. Analysis and synthesis filter bank.
filter bank coefficients are not the same. Since the waveform of the single note of the musi-
cal signal which is used as the scaling function does not satisfy the orthogonality property
they are designed to be biorthogonal wavelets. Hence the remaining analysis and synthesis
filter bank coefficients can be found out using the properties of biorthogonal wavelet. Using
the biorthogonal wavelet and scaling function coefficients, faithful reconstruction can be
obtained. Inorder to get more accurate results the the filter bank coefficients can be modified
using lifting technique [16]. Lifting results in a new wavelet with enough vanishing moments.
Denoising of noisy signals can be seen as an important application of any wavelet designed
specifically for that signal. Wavelet based denoising is done by the thresholding of wavelet
coefficients. In wavelet analysis high amplitude coefficients mainly represent signal and low
amplitude coefficients with randomness represent noise. If a signal has its energy concentrated
in a small number of wavelet coefficients, its coefficients will be large compared to any other
signal or noise that has its energy spread over a large number of coefficients. Denoising is
achieved by selecting an appropriate threshold for such high amplitude coefficients [17]–[18].
The rest of the paper is organised as follows. Section II briefly explains filter bank theory.
Section III explains the need for biorthogonal wavelets and their design methodology. The
method of modifying the wavelets using lifting technology is given in section IV. Section V
explains the denoising of musical signals. The simulation results are given in Section VI. The
paper is concluded in Section VII.
2 FILTER BANK THEORY
Wavelet decomposition and reconstruction of a musical signal based on the multi resolution
theory can be obtained using digital FIR. Figure 2 shows the filter bank implementation
of wavelets. The filters h(n) and g(n) are analysis (decomposition) filters which are low pass
and high pass respectively. They are downsampled by two. The high frequency coefficients
are detailed coefficients and low frequency coefficients are approximation coefficients. Since
musical signals need biorthogonal wavelet, the synthesis reconstruction filters are dual of the
analysis filters. They are denoted as h( n ) and g ( n ) which are preceded by upsampling by
two to give the reconstructed signal.
3 DESIGN OF BIORTHOGONAL WAVELETS
In the case of musical signals, the repeating elements are such that the scaling function
obtained using them does not satisfy the orthogonal condition. They therefore need to be
designed as biorthogonal. The analysis and synthesis filter coefficients are different in the
biorthogonal case. In biorthogonal wavelets, there is a dual scaling function in addition to the
scaling function generated by h(n). It is denoted by φ(t ). In the case of orthogonal wavelets
f(t) is orthogonal to its own translates whereas in the case of biorthogonal wavelets, f(t) is
orthogonal to the translates of φ(t ) . Similarly, f(t) is orthogonal to ψ(t) for ordinary wave-
lets. But f(t) is orthogonal to ψ (t ) for biorthogonal wavelets.
Mathematically biorthogonal wavelets imply,
673

φ (t ) = ∑ h( n )φ (2t − n ) (4)
n
φ(t ) = ∑ h( n )φ(2t − n ) (5)

n
ψ (t ) = ∑ g ( n )φ (2t − n ) (6)
n
ψ (t ) = ∑ g ( n )φ(2t − n ) (7)
n
where
g ( n ) = ( −1)n h(1 − n ) (8)
and
g ( n ) = ( −1)n h(1 − n ) (9)
The filter coefficients h(n), are obtained using MPSO algorithm. The coefficients for the dual
scaling function, h( n ) , are designed from h(n) so that they satisfy the following conditions.
Normality of the dual scaling function:-
∑ h( k ) = 2 ∀ k ∈ Z
k
(10)
f(t) should be orthogonal to the translates of φ(t ) :-
∑ h(k )h( k − 2n) = δ

k
n,0 ∀ k ∈Z (11)
First vanishing moment of ψ(t):-

∞
∫ ψ (t )dt = 0 ⇒ ∑ ( −1)n h( N − k − 1) = 0 (12)
−∞
k
h( n ) is obtained by solving the Equation (10), Equation (11) and Equation (12). The obtained
h(n) and h( n ) values are used to design g(n) and g ( n ) as shown in Equation (8) and Equation
(9). These values are substituted in Equation (6) and Equation (7) to get the biorthogonal
wavelet functions.
4 LIFTING TECHNOLOGY
The lifting scheme is an approach to construct so called second generation wavelets, which
are not necessarily transalates and dilations of one function [16]. In the present work, lifting
scheme in the Z domain is used for making new set of filter coefficients from the existing
wavelet. It works in spatial domain. Using the lifting technology, a set of filter coefficients
can be modified into a new set of filters without affecting the perfect reconstruction property.
Figure 3 shows the basic idea of lifting. Filter coefficients u(n) in the left part of the diagram
modifies the high pass filtered signal by adding it a weighted sum of low pass filtered signal
coefficient. On the right part of the diagram u(n) nullifies this change by subtracting the same
quantity. In the left part of the diagram u(n) modifies the detail coefficients there by modify-
ing high-pass analysis filter coefficients. In the right side u(n) performs the undo operation
giving the highpass filtered signal back. This results in a set of new analysis low-pass filter
coefficients and synthesis highpass filter coefficients. Lifting of SSM wavelets is described as
follows.
674

Figure 3. Analysis and synthesis filter bank using lifting.
Initially h(n) and g ( n ) are taken as such. g(n) and h( n ) are taken as Haar wavelet coef-
ficients. ie, g(n) = [1 −1] and h( n ) = [1] . In the Z domain, the four filter coefficients can be
written as
H ( z ) = h1 ( n ) + h2 ( n )z −1 + … + hK ( n )z − K (13)
−1
G(z) = 1 − z (14)
H ( z ) = 1 + z −1 (15)
G ( z ) = g1 ( n ) + g2 ( n )z −1 + … + gK ( n )z − K (16)
Suppose new wavelet function at the analysis side is
ψ new (t ) = ψ (t ) + α1φ (t ) + α 2φ (t − 1) (17)
Applying the moment conditions to above equations, the constants α1 and α2 can be
found out and the new wavelet function is obtained. From this modified analysis highpass
filter, gnew(n) can be obtained. The equation connecting gnew(n) and u(n) in Z domain is as
follows:
G new ( z −1 ) = G ( z −1 ) + U ( z 2 )H ( z −1 ) (18)
u(n) can be found out from above equation so that modified h new ( n ) can be obtained as
follows:
H new ( z ) = H ( z ) − U ( z 2 )G ( z ) (19)
5 DENOISING USING WAVELETS
Consider a clean musical signal x(n) of length N is corrupted by a additive white gaussian
noise denoted by w(n). Then the noisy signal v(n) is given by
v( n ) = x ( n ) + ω ( n ) (20)
Here the idea is to recover back the signal x(n) from this noisy signal. For a particular
musical instrument, the most suited wavelet will have maximum energy concentrated in
the approximation coefficients rather than in the detailed coefficients. Hence only detailed
coefficients are denoised. The denoising is done by proper thresholding of detailed coefficients
obtained after decomposition of the noisy signal. Hence denoising can be viewed as three
steps.
1. Decomposition of the noisy signal
2. Thresholding of the Detailed Coefficients
3. Reconstruction of the original signal
675

5.1 Estimation of threshold
In this paper denoising is done by adopting the universal threshold proposed by Donoho
et al. [17]. It is described as follows:
If N is the length of the signal the threshold λ at level j is given by
λ j = σ j 2 log( N ) / N (21)
where the scale estimate σj is given by
σ j = MAD / 0.6745 (22)
where MAD is the median absolute value of the wavelet coefficients at level j.
5.2 Thresholding of detail coefficients

If the noise energy is less than signal energy, the corresponding noise wavelet coefficients
will be obviously less than the signal wavelet coefficients. By soft thresholding, the detail
coeffficients can be thresholded. If vj,k is the wavelet coefficients at level j, soft thresholding
is done by,
 υ − λ , if | υ j ,k | ≥ λ
υ j ,k =  j ,k (23)
 0, otherwise
6 SIMULATION RESULTS
6.1 Biorthogonal SSM filter coefficients

Once h(n) is obtained using modified MPSO algorithm, h( n ) is obtained by solving equations
formed by substituting different integer values for k in Equations (10)–(12). The coefficients,
h( n ), are symmetrical and also the length is more than that of h(n). The length of h( n ) for
flute signal is 17 while that of h(n) is 15. The coefficients obtained are as shown in the Table 1.
It can be seen that h(n) and h( n ) satisfy all the Eqns. (10)–(12). From these values, g(n) and
g ( n ) are obtained using Equation (8) and Equation (9). The original signal and the recon-
structed signal using biorthogonal SSM wavelet is shown in the Figure 4.
Table 1. Filter bank coefficients for flute signal.
h(n) 0.2290 0.1796 0.0235 0.2084 0.2928

0.1962 -0.0004 -0.0201 0.1703 0.2593
-0.0139 -0.1915 0.0359 0.0728 -0.0244
g(n) -2.1465 -2.9536 -0.5659 -1.4113 0.4386
-2.9206 0.6092 6.9320 4.0364 6.9320
0.6092 -2.9206 0.4386 -1.4113 -0.5659
-2.9536 -2.1465
n)
h( -2.1465 2.9536 -0.5659 1.4113 0.4386
2.9206 0.6092 -6.9320 4.0364 -6.9320
0.6092 2.9206 0.4386 1.4113 -0.5659
2.9536 -2.1465
n)
g( -0.0244 -0.0728 0.0359 0.1915 -0.0139
-0.2593 0.1703 0.0201 -0.004 -0.1962
0.2928 -0.2084 0.0235 -0.1796 0.2290
676

Figure 4. Reconstructed signal using biorthogonal SSM wavelet.
Table 2. Filter bank coefficients modified using lifting.
h(n) 0.2290 0.1796 0.0235 0.2084 0.2928

0.1962 -0.0004 -0.0201 0.1703 0.2593
-0.0139 -0.1915 0.0359 0.0728 -0.0244
g(n) 0.9531 -1.036 0.042 -0.00058 -0.055
0.00025 0.06 0.04427 -0.35 -0.0571
0.0377 0.0922 -0.00045 -0.0541 0.0123
0.0728 -0.0244
n)
h( 1.0236 1.7047 -0.0103 -0.1122 -0.02245
0.0597 -0.1509 0.2399 -0.1702 0.1698
-0.283 0.3979 -0.3155 0.3822 -0.2451
0.1796 -0.229
n)
g( -0.0244 -0.0728 0.0359 0.1915 -0.0139
-0.2593 0.1703 0.0201 -0.004 -0.1962
0.2928 -0.2084 0.0235 -0.1796 0.2290
6.2 Modified SSM filter coefficients after lifting

The biorthogonal filter coefficients obtained above are modified using lifting technology as
described before. The modified filter coefficients are as shown in the Table 2. The length
of the filter coefficients are same as before. The reconstructed signal using modified SSM
wavelets after lifting is shown in the Fig. 5. The signal is more similar to the original signal
compared to Fig. 4.
6.3 Denoising of flute signals
1. Using standard wavelets: The set of flute signals are denoised using standard wavelets.
Three wavelets families, Symlets 2 to 8, Daubechies 2 to 10 and Coiflet 1 to 5 are taken.
The result shows that Coiflet5 gives best denoising performance.
2. Using biorthogonal SSM wavelets: The set of flute signals are denoised using biorthogonal
SSM wavelets. The flute signal corrupted with noise is first decomposed to approximation
and detail coefficients. The threshold is calculated as per Eqn. (21) and Equation (22).
The detail coefficients are thresholded by using equation (23). Finally the signal is recon-
structed back. Figure 6 shows the result.
677

Figure 5. Reconstructed signal using biorthogonal SSM wavelet after lifting.
Figure 6. Denoised signal using biorthogonal SSM wavelet.
Figure 7. Denoised signal using modified biorthogonal SSM wavelet.
3. Using modified SSM wavelets after lifting: The set of flute signals are denoised using
modified SSM wavelets using the same method described above. Figure 7 shows the result.
The denoised signal is more similar to the original signal compared to the result in Figure 6.
678

Table 3. SNR comparison in DB.
Wavelet Input SNR Output SNR
Coiflet5 20.60 26.93

Coiflet1 20.60 24.48
Haar 20.60 23.97
daubechies10 20.60 26.52
Sym1et8 20.60 26.46
Symlet1 20.60 23.92
Biorthogonal SSM 20.60 24.52
Modified SSM 20.60 29.03
7 COMPARISON OF DENOISING PERFORMANCE
In order to compare the quality and performance of the denoised signal using two set of
filter coefficients, the signal to noise ratio of the signal is determined. SNR is calculated as
follows.
 N −1

 ∑ xk2 
SNR = 10 log  N −1 k = 0  (24)
 2
∑ k ( x − x k ) 
 k =0 
where x is the original signal and x is the denoised signal. The value of SNR obtained are as
shown in the Table 3. Modified SSM wavelet gives the highest SNR.
8 CONCLUSION
Musical signals are analyzed using a wavelet specifically designed for these signals. Using the
already available analysis low-pass coefficients of this wavelet, biorthogonal SSM analysis
and synthesis filter coefficients are found out by adopting the properties satisfied by the
biorthogonal wavelets. Using these coefficients, the musical signal is decomposed and recon-
structed. The coefficients are modified using lifting technology with an aim to improve the
reconstruction performance. The resulting reconstructed signal shows more similarity to the
original signal than the previous one. Denoising of musical signals is performed using these
filter bank coefficients. An appropriate threshold for removing the detail coefficients corre-
sponding to noise is calculated. The noisy signal is denoised using standard wavelets. Among
the standard wavelets, Coiflet5 is found to be the suitable wavelet for denoising a musical
signal. It is able to reduce the noise in the noisy musical signal when it is denoised using the
two sets of SSM coefficients obtained. The result shows that applying lifting improves the
denoising performance.
REFERENCES
[1] I. Daubechies. Ten Lectures on Wavelets. SIAM, Philadelphia, PA, 1992.

[2] Hubbard, Barbara Burke, “The World According to Wavelets The Story of a Mathematical Technique
in the Making”, Universities Press, 1998.
[3] Martin Vetterli and Cormac Harley, “Wavelets and filter banks: Theory and design”, IEEE
Transactions on Signal Processing, Vol. 40, No. 9, September 1992, pp. 2207–2232.
[4] Joseph.O. Chapa and Raghuveer M.Rao, “Algorithms for designing wavelets to match a specified
signal”, IEEE Transactions on Signal Processing, Vol. 48, No. 12, December 2000, pp. 3395–3406.
679

[5] J. Nagumo and A. Noda “A learning method for system identification”, IEEE Transactions on
Automatic Control, Vol. AC-12, June 1967, pp. 283–287.
[6] Markus Rupp, The Behaviour of LMS and NLMS Algorithms in the Presence of Spherically Invari-
ant Process, IEEE Transactions on Signal Processing. Vol. 41, No. 3, pp. 1149–1160, March 1993.
[7] Raghavendra Sharma and V Prem Pyara, “A Comparative Analysis Of Mean Square Error Adap-
tive Filter Algorithms For Generation Of Modified Scaling And Wavelet Function”, International
Journal Of Engineering Science And Technology (IJEST), Vol. 4, No. 4, April 2012, pp. 1402–1407.
[8] Jyoti Dhiman, Shadab Ahmad and Kuldeep Gulia, “Comparison Between Adaptive Filter Algo-
rithms (LMS, NLMS and RLS)”, International Journal Of Science, Engineering And Technology
Research (IJSETR), Volume 2, Issue 5, May 2013, pp. 1100–1103.
[9] Thamer M. Jamel, “Performance Enhancement of Adaptive Acoustic Echo Canceller Using a New
Time Varying Step Size LMS Algorithm (NVSSLMS)”, International Journal of Advancements
in Computing Technology(IJACT), Korea, Vol. 3, No. 1, January 2013.
[10] R.H. Kwong and E.W. Johnston, “A variable step size LMS algorithm”, IEEE Transactions on
Signal Processing, Vol. 40, July 1992, pp. 1633–1642.
[11] Junghsi Lee, Jia-Wei Chen, and Hsu-Chang Huang, “Performance Comparison of Variable Step-
Size NLMS Algorithms”, Proceedings of the World Congress on Engineering and Computer Sci-
ence 2009, Vol. 1, October 2009.
[12] M.S. Sinith, Madhavi N. Nair, Niveditha P. Nair and Parvathy S. Identification of Wavelets and
Filter Bank Coefficients in Musical Instruments, International Conference in Audio Language and
Image Processing (ICALIP 2010), Shanghai, China pp. 727–731, Nov. 2010.
[13] J Kennedy and R Eberhart, Particle swarm optimization, Proceedings of IEEE International Con-
ference on Neural Networks, vol. 4, pp. 1942–1948, 1995.
[14] R Poli, J Kennedy and T Blackwell, Particle swarm optimization, International Journal of Swarm
Intelligence, Volume 1, Issue 1, pp. 33–57, June 2007.
[15] M.S. Sinith, Shikha Tripathi, K.V.V. Murthy, “SSM wavelets for analysis of music signals using
particle swarm optimization”, IEEE International Conference on Signal Processing and Communi-
cation (ICSC), Noida, India, pp. 247–251, Dec. 2013.
[16] K.P. Soman and K.I. Ramachandran, “Insight into Wavelets:From Theory to Practise”, Prentice
Hall.
[17] Donoho D.L., Johnston I.M., “De-noising by soft-thresholding” IEEE transactions on information
theory, vol. 41, No. 3, 1995, pp. 613–627.
[18] Donoho D.L., Johnston I.M., “Ideal spatial adaptation by wavelet shrinkage.” Biometrika, Vol. 81,
No. 3, 1994, pp. 425–455.
680

Low complexity encoding of M-ary QAM constellation

for linear index codes
Anjaly Shaju & Senthilkumar Dhanasekaran

Department of Electronics and Communication, Ramaiah Institute of Technology, Bangalore, India
ABSTRACT: An index coding problem consists of a single source S with K messages

transmitted across a Gaussian broadcast channel, where each receiver demands a set of
information from the source while having a subset of prior information in its cache, known
as ‘side information’. The power of the receiver which efficiently exploits receivers’ side
information is called the ‘side information coding gain’. The known index codes have the
heaviest quantity of side information gain, but finding the encoding matrix is a tedious process.
This presented work aims to find the encoding matrix to construct the multidimensional
Quadrature Amplitude Modulation (QAM) constellation proposed by Natarajan et al.
(2015a), with a fewer number of computer searches. Furthermore, the side information gain
achieved is compared with the existing method.
1 INTRODUCTION
Network coding is a promising technology which offers advantages to the communication

networks, such as throughput and performance of the network (Koetter & Médard 2003).
In a multicast communication network, source nodes broadcast data to the intermediate
nodes. The intermediate nodes combines the data from the source nodes and broadcast to the
destination nodes. Hence, it helps in minimizing the bandwidth and the delay.
An equivalence between the network coding and index coding was proved by Effros et al.
(2015). Any network coding problem can be converted into the index coding problem, the
solution obtained to the index coding problem, and then the solution returned back into the
original network coding problem. Index coding is the reduced form of the network coding.
The equivalence between these codes holds good for both the linear and nonlinear codes.
Index coding was first proposed (Birk & Kol 2006) for satellite communication, where a
single source S with a set of receivers (R1, R2, …, Rn) are considered and each receiver has
some subset of information as a priori. Information Source Coding on Demand (ISCOD)
entails exploiting full knowledge of client side information to fulfill the receivers’ demand.
In satellite communication, during the main transmission, all the clients may not satisfy its
demand due to insufficient storage capacity in the client’s cache or due to any other interfer-
ences. There is no direct communication between the clients, but a feedback channel is used
to send the requests back to the server. In this case, index coding is used in wireless networks
to satisfy the receivers’ demand with a fewer number of binary transmissions. Optimal index
codes can be used to reduce the number of binary transmissions by using the min-rank (Bar-
Yossef et al. 2011, Mahesh and Rajan 2016). The optimal index codes is constructed by using
the encoding matrix, which consists of receiver-side information and its demand.
A system model for lattice index codes (Natarajan et al. 2015b) consists of a single source
S and a set of K messages, where each receiver demands some subset of messages from the
source while having some information in its cache as a priori. Lattice index codes efficiently
exploit the receiver-side information and convert it into apparent coding gain. The lattice
index codes for the Gaussian broadcast channel, in which the K messages are individually
mapped into the K modulo lattice constellation and the transmitting symbols are generated
681

by taking the sum of the individual symbols. Natarajan et al. (2015a) proposed index codes
based on a multidimensional Quadrature Amplitude Modulation (QAM) constellation for
the Gaussian broadcast channel, where every receiver demands all the messages from the
source. Side information coding gain is obtained by efficiently exploiting the receiver-side
information present in its cache, over a non-fading broadcast channel with K independent
messages. The known index codes have larger complexity in the construction of the encoding
matrix, which is uniquely decodable.
The noisy index coding (Natarajan et al. 2015b, Natarajan et al. 2015a) is a specific case
of index coding where all the receivers demands K independent messages from the source,
indicating (w1, w2, …, wk) that assumes the value from (W1, W2 , …, WK ). The lattice codes
(Natarajan et al., 2015b) are constructed by using the Chinese remainder theorem with larger
side information gain. The K messages are individually mapped into the K modulo lattice
constellation and the transmitting symbols are generated by taking the sum of the individual
symbols. The source is assumed to be operated in average power constraints and the receiver
experiences Additive White Gaussian Noise (AWGN). Each receiver is denoted by (SNR, S),
where SNR denotes the signal-to-noise ratio and S = 1, …, K where S represents the side
information present in each receiver.
The terminology S = f indicates that there is no side information present in the receivers.
Let (R1, R2, …, RK) be the rate of the each messages in bits per dimension. Suppose the source
entropy is R = (R1, R2, …, RK), and the side information rate is represented by RS = ∑ K ∈S RK
(Natarajan et al. 2015b). The index code presented by Natarajan et al. (2015b) offers larger
receiver side information coding gains, but there are two practical limitations: i) It does not
encode all the messages at equal rates; and ii) It does not consider the messages that are
a power of two. Recently, Natarajan et al. (2015a) have proposed multidimensional QAM
constellation mapping for index codes over the Gaussian broadcast channel, where all the
receivers demand an equal number of messages from the source. With the help of computer
searches, an encoding matrix is constructed with the symbols in the set ZM, with the deter-
minant as the odd integer in the set, and up to five messages with message size of 2m ≤ 6.
The complexity is exponentially increased in the construction of the encoding matrix when a
larger number of messages are considered.
In this paper, we proposed an algorithm which reduces the computer searches, in turn,
minimizing the complexity in the construction of the encoding matrix to map the symbols on
the multidimensional QAM constellation. Furthermore, we have shown that the side infor-
mation coding gain achieved by the code for a different number of side information is known
at the receiver. The obtained results are compared with the recently proposed algorithm,
which clearly shows that the obtained encoding matrix offers better side information gain in
some cases but inferior in some other scenarios.
2 CONSTRUCTION OF ENCODING MATRIX FOR QAM CONSTELLATION
In this section, we discuss an algorithm to construct the encoding matrix which linearly trans-
forms the K-tuple symbols from the ZMK message symbol space into an ZMK coded symbol
space. This transformation includes all the possible K-tuple symbols in the construction of
the transformation matrix (i.e., circulant encoding matrix). For example, K-tuple symbols
with the symbol set size of M, there are MK possible encoding matrices. Hence, this will expo-
nentially increase the complexity with K, in the construction of an optimum encoding matrix
to map the message symbols onto the coded symbols. The simplest way to find K linearly
independent code vectors is to construct the circulant matrices using all possible symbols,
which significantly reduces the complexity of the computer searches (Natarajan et al. 2015a).
In order to span the K-tuple message symbol space over modulo M, we consider a
{ }
symbol set, ZM − 2M , − ( M2 − 2 ) , …, 0, …, ( M2− 2 ) for even values of M, and for odd values of M,
{ }
ZM − ( M2 −1) , − ( M2 − 3 ) , …, 0, …, ( M2−1) . The ZM has the structure of the commutative ring with
addition and multiplication performed over an integer modulo M. A unit, U(M), of the
defined set ZM, is the odd integers in the set ZM when M is even, and even integers when M is
682

odd. A linear index code with K messages consists of a set of K generators where the linear
encoder X = ρ (w1, w2 ,… wK ) = ∑ k =1 wk ck mod M is injective.
K
The injectivity of the linear encoder X gives the unique decodability at the receiver side. A lin-
ear index code is completely characterized by the matrix C whose rows are K generators c1, c2, …,
cK. The encoding matrix C defines a linear transformation in which the matrix multiplies with
message symbols to form the codewords. Thus the encoder mapping is injective if and only if C
is invertible. The matrix is constructed with the determinant as the unit of the symbol set ZM.
3 SIMPLIFIED ALGORITHM TO CONSTRUCT ENCODING MATRIX
3.1 Proposed algorithm

In this section, we propose an algorithm as shown in Figure 1, which minimizes the complexity
in the computer searches significantly, to construct the encoding matrix. We consider the
following issues to frame a simplified algorithm: i) Exchanging of the rows in the circulant
matrices will provide the same coding gain; and ii) An odd number of odd integers in the first
row of the circulant matrix will provide the determinant value of an odd integer.
Remark 1 Circulant matrix which has an odd number of odd integers in the first row, will
offer the determinant value of an odd integer.
Example 1. For a 2 × 2 matrix, the odd number of odd integers in the first row of the circulant
encoding matrix gives an odd integer as the determinant.
o1 e1
= o1.o1 − e1.e1 = o − e = odd number (1)
e1 o1
Similarly, for a 3 × 3 matrix:
o1 e1 e2
e2 o1 e1 = odd number (2)
e1 e2 o1
Figure 1. Flowchart for simplified algorithm to construct the encoding matrix.
683

Algorithm
Step 1: Construct the symbol set ZM
Step 2: Construct all possible K-tuple symbols
Step 3: Construct the circulant matrix
Step 4: If cyclic shift of that matrix already exists, repeat step 2
Step 5: If even number of odd integers in the K-tuple symbols, repeat step 2
Step 6: Stack the symbols to construct the encoding matrix
Example 2. Let K = 2 and M = 8. The symbol sets are mapped into the set ZM to form a com-
mutative ring. For K = 2, we construct the 2 × 2 encoding matrix, which is circulant, and the
symbols with an odd number of odd integers are chosen as the first row to construct the cir-
culant matrix. The encoding matrix should satisfy the conditions such that the determinant
should be in the unit of the ring.
For K = 2 and M = 8 the two generators are c1 = (−2, −3) and c2 = (−3, −2). The encoder
x = (−2w1 −3w2, −3w2 −2w1) mod 8, is calculated from the rows of the circulant encoding
matrix. The circulant encoding matrix for K = 2 and M = 8 is given by:
 −2 −3 
C= 
 −3 −2 
The determinant for the encoding matrix is, −5 mod 8 = −1, which is in the unit of the set
ZMK . Therefore, this linear encoder is injective. The labelling scheme for a two dimensional
64- QAM constellation is shown in Figure 2.
3.2 Complexity
In an existing algorithm (Natarajan et al. 2015a), in order to reduce the complexity in the
exhaustive search space, the authors have chosen the circulant matrices whose determinant is
in unit of a set ZM as the encoding matrices, which provides the largest minimum Euclidean
distance between the coded symbols. As discussed in Section 3.1, the proposed algorithm
minimizes the complexity by reducing the computer searches significantly. For example, for
K = 2 and M = 32, there are 322 = 1024 possible symbols (i.e., encoding matrices). The number
of searches, computed from the simulation, in the proposed algorithm is 257, whereas in
(Natarajan et al. 2015a) it is, K × 257 = 2 × 257 = 514.
Figure 2. The 64-QAM constellation. The eight points forming the subcode corresponding to the side
information when w1 = −4 are highlighted with circles and the subcode for w2 = −4 is marked with squares.
684

3.3 Side information gain
The ability of the receiver to efficiently exploit the receiver-side information is called ‘side
information coding gain’. It is assumed that all K messages have equal transmission rate. The
transmission rate is given by:
1
Rk = log 2 M b/dim (3)
K
The side information rate at each receiver is given by:
|S |
RS = ∑R
k∈S
k =
K
log 2 M b/dim (4)
The coding gain achieved with side information available at the receiver, is given by:
10 log10 d s2 / d02
Γ dB/b/dim (5)
RS
Table 1. Comparison between the existing algorithm and the proposed algorithm (one side informa-
tion is known) with first row of the circulant encoding matrix and the gain.
Algorithm in (Natarajan et al. 2015a) Proposed Algorithm
M K = 2 K = 3 K = 4 K = 2 K = 3 K = 4
4 (1,-2) (1,-2,-2) (1,1,-1,0) (-1,-2) (-2,1,-2) (-2,-2,-2,-1)

6.02 4.515 6.02 6.02 4.515 6.02
8 (1,2) (1,2,0) (1,0,3,3) (-2,-3) (3,-2,0) (-4,-1,2,2)
4.65 6.02 4.01 4.65 6.02 4.01
16 (1,-4) (1,2,-6) (1,4,-6,-8) (-3,-4) (2,-6,7) (-8,-5,-4,2)
6.02 5.24 6.02 6.02 5.24 6.02
32 (1,6) (1,-10,14) (1,10,14,2) (-10,-9) (-15,-10,14) (-15,10,14,2)
5.85 5.73 6.22 5.85 5.73 6.22
64 (1,-28) (1,-26,-4) (1,-26,20,30) (-11,-12) (-20,8,17) (-31,-30,20,26)
6.04 5.73 6.36 6.04 5.57 6.36
Table 2. Comparison between the existing algorithm and the proposed algorithm (two side informa-
tion are known) with first row of the circulant encoding matrix and the gain.
M K = 3 K = 4 K = 3 K = 4
4 (1,-2,-2) (1,1,-1,0) (-2,1,-2) (-2,-2,-2,-1)

4.515 3.01 4.515 3.01
8 (1,2,0) (1,0,3,3) (3,-2,0) (-4,-1,2,2)
3.49 5.18 3.49 4.65
16 (1,2,-6) (1,4,-6,-8) (2,-6,7) (-8,-5,-4,2)
5.83 5.57 5.83 5.57
32 (1,-10,14) (1,10,14,2) (-15,-10,14) (-15,10,14,2)
5.84 6.02 5.84 6.02
64 (1,-26,-4) (1,-26,20,30) (-20,8,17) (-31,-30,20,26)
5.815 5.85 6.02 5.85
685

Table 3. Comparison between the existing algorithm and the proposed algorithm (three
side information are known) with first row of the circulant encoding matrix and the gain.
M K = 4 K = 4
4 (1,1,-1,0) (-2,-2,-2,-1)
3.18 4.01
8 (1,0,3,3) (-4,-1,2,2)
4.62 5.35
16 (1,4,-6,-8) (-8,-5,-4,2)
6.02 6.02
32 (1,10,14,2) (-15,10,14,2)
5.80 5.80
64 (1,-26,20,30) (-31,-30,20,26)
6.08 6.08
where dS is the distance between the constellation points when either of the side information is
known to the receiver and d0 is the distance between any two adjacent points in the constellation.
In this section, we present the results for K = 2, 3, 4 and M = 4, 8, 16, 32, 64. With the help of
a computer search we find the best linear index codes which provide maximum coding gain.
The coding gains achieved by the code for one, two, and three side information are known,
and respectively, detailed in Tables 1, 2, and 3. The results presented in Tables substantiate
that the encoding matrix suggested by the proposed algorithm offers the comparable side
information gain compared to the existing work, with low computational complexity.
5 CONCLUSION
The construction of an encoding matrix for multidimensional QAM constellation mapping

of linear index codes is presented. Moreover, it is substantiated that the proposed simplified
algorithm significantly reduces the computer searches, and in turn, the complexity in the con-
struction of the encoding matrix for index codes is significantly less, compared to the existing
algorithm (Natarajan et al. 2015a). Simulation results presented in Tables have shown that the
proposed algorithm performs similar to its counterpart (Natarajan et al. 2015a), with low com-
plexity. It is shown that the complexity in the proposed algorithm is reduced by a factor of K.
REFERENCES
Bar-Yossef, Z., Y. Birk, T. Jayram, & T. Kol (2011). Index coding with side information. IEEE Transac-
tions on Information Theory 57(3), 1479–1494.
Birk, Y. & T. Kol (2006). Coding on demand by an informed source (ISCOD) for efficient broadcast of dif-
ferent supplemental data to caching clients. IEEE/ACM Transactions on Networking 14(5), 2825–2830.
Effros, M., S. El Rouayheb, & M. Langberg (2015). An equivalence between network coding and index
coding. IEEE Transactions on Information Theory 61(5), 2478–2487.
Koetter, R. & M. Médard (2003). An algebraic approach to network coding. IEEE/ACM Transactions
on Networking 11(5), 782–795.
Mahesh, A.A. & B.S. Rajan (2016). Noisy index coding with PSK and QAM. arXiv preprint
arXiv:1603.03152.
Natarajan, L., Y. Hong, & E. Viterbo (2015a). Index codes for the gaussian broadcast channel using
quadrature amplitude modulation. IEEE Communications Letters 19(8), 1291–1294.
Natarajan, L., Y. Hong, & E. Viterbo (2015b). Lattice index coding. IEEE Transactions on Information
Theory 61(12), 6505–6525.
686

Improving myoelectric grasp recognition using empirical mode

decomposition and differential evolution based approach
C.K. Anusha & K. AjalBabu

Department of Electrical and Electronics Engineering, TKM College of Engineering, Kollam, India
Nissan Kunju
Department of Electronics and Communication Engineering, TKM College of Engineering, Kollam, India
ABSTRACT: This paper presents a surface Electromyographic (sEMG) signal based

hand grasp recognition technique utilizing Empirical Mode Decomposition (EMD) and
Differential Evolution based Feature Selection (DEFS). A series of features were derived
from both the raw signal and its corresponding Intrinsic Mode Functions (IMF’s), obtained
by performing EMD. Differential Evolution (DE) is a relatively new soft computing tech-
nique with wide range of applications. Being a promising stochastic population based opti-
mization method, a feature selection framework using DE is utilized in this research to
identify the optimum feature subset. sEMG signals recorded from eleven healthy subjects
are used for this study. The proposed method is further validated using other popular feature
extraction techniques and different pattern recognition algorithms. Outcome of our research
shows that the methodology of using EMD with DEFS can achieve significant improvement
in the overall recognition rate.
1 INTRODUCTION
Assistive devices for neurological rehabilitation, for example active prostheses, are controlled
by man machine interfacing. Nowadays myoelectric control is evolved as the most promising
approach to control devices utilized in clinical and commercial applications (Jiang et al.2012,
Fougner et al. 2012, Scheme & Englehart 2009). In spite of the fact that nerve and brain
recording are exceptionally encouraging for a direct neural interfacing, they often require
invasive methods for electrode placement that limits their practical applicability to labora-
tory research or small-scale clinical testing (Micerra & Navarro 2009). Although Industrial
developers like Otto Bock (Germany) and Touch Bionics (USA) have introduced surface
EMG based artificial limbs in the market, EMG based control is still in a premature state
being limited to few hand postures and higher EMG-channels required for effective control.
In this paper a scheme for classification of human hand grasps from surface EMG signals
is presented. The novelty of our approach dwelled on the use of EMD for feature extrac-
tion combined with Differential Evolution Based Feature Selection. The feature selection
framework that have been utilized in this study also gives a versatile approach to improve
the developed models comprehensibility by selecting the optimum feature subset adaptively
(Khushaba R. et al. 2008, Storn R. 2008, Ahmed Al-Ani et al. 2013). Our results prove that
the methodology of using EMD with DEFS can achieve significantly good results.
2 PROPOSED METHODOLOGY
The work presented in this paper stems from the desire to design a self-contained pros-
thetic system. For the laboratory stage of the work a standard PC installed with windows
687

OS is used. EMG recordings were made from a generic Data Acquisition System-CMCdaq
(Nissan Kunju et al. 2013). EMG recordings were processed offline using Matlab 2014(a)
software. The extracted signal is bandpass filtered and non-contracting portions at the begin-
ning of the recording are effectively removed by carrying out signal thresholding. The sEMG
signal after preprocessing is segmented using an overlapping windowing approach and vari-
ous features were extracted both from the raw signal and its corresponding intrinsic mode
functions (IMF’s), derived by performing EMD. A feature selection stage using Differential
Evolution Algorithm is employed after feature extraction to avoid the effects of dimensional-
ity. Pattern recognition phase uses the extracted features and classify each segment to one of
the six hand grasps.
3 ELECTROMYOGRAM ACQUISITION
The EMG data is recorded from eleven healthy subjects (aged between 20–30 years). Before
the start of the experiment, subjects were thoroughly familiarized with the experimental
protocol and the EMG equipment. A four channel generic EMG data acquisition system
CMCdaq is used to acquire the data at a sampling rate of 1000 Hz. The Ag/AgCl electrodes
were attached over the muscle belly in line with the muscle fibres in accordance with the
standard procedure in literature (Shrirao N.A. et al. 2009). The four surface EMG electrodes
were placed on forearm muscles Flexor Capri Ulnaris, Extensor Capri Radialis, Extensor
Digitorum, Flexor Digitorum Superficialis and ground electrode was placed onthe contralat-
eral upper limb. This follows the electrode placement discussed in (Frank F.H. 1989). Three
trials of the six different grasps, each with a duration of six second was performed and the
speed and force was intentionally left to the subject’s will. The Maximum Voluntary Isomet-
ric Contraction (MVIC) test was also executed by having the subject to flex and extend his/
her hand at the wrist joint by exerting maximum possible force to the maximum possible
inclination and sustaining it up to six seconds. The six basic hand grasps (Schlesinger G.
1919) and the experimental setup is shown in Fig. 1 and Fig. 2 respectively.
Figure 1. Illustration of basic hand grasps.
688

Figure 3. Illustration of evolution of the MAV values of EMG signal starting from rest along with the
threshold value (shown as dotted line) for spherical grasp.
4 SIGNAL PRE PROCESSING
Surface Electromyographic (sEMG) signals are usually affected by noise and in practice, the
acquired signal may be corrupted hence the raw signal need to be preprocessed before further
processing. The usable frequency range of EMG signal is considered to be in the range of
15–500 Hz as most of the energy is concentrated at this specified frequency range (De Luca
C.J 1998). The acquired signal is filtered using a 4th order butterworth bandpass filter with
high pass cut off frequency at 15 Hz and low pass cut off frequency at 500 Hz. After filtering
each channel is normalized using the MVIC obtained for each muscle.
For each grasp the level of involvement of each muscle will be different and a dynamic
threshold selection approach based on local characteristics of the signal is implemented
for offline analysis (Fig. 3). Sliding window approach is used to focus only on segments
where muscle is contracted. Keeping sliding window size as 50 ms, Mean Absolute Value
(MAV) is calculated for each window and once that value exceeds a threshold the muscle
is no longer considered to be in resting phase. In order to preserve the temporal informa-
tion, recordings were taken for processing on due activation of any one of the four chan-
689

nels. Even though this thresholding cannot be used directly in online validation since it
requires prior knowledge about the signal, this scheme worked out well in our offline
analysis.
5 FEATURE EXTRACTION AND SELECTION
Each grasping operation is characterized by a unique motor unit firing pattern (Kilbreath
et al. 2002, Boser B.E. et al. 1992), identifying and describing patterns that better discriminate
different classes of grasp over different trials. This is the core philosophy of feature extrac-
tion. The adaptive nature of decomposition and its ability to preserve the varying frequency
in time makes EMD a powerful choice in the analysis of EMG signals (Huang et al. 1998,
Andrade A.O. 2004, Flandrin P. 2004). In this study 10 most popular features (Ericka Janet
& Housheng Hu 2011) from time and frequency domain are used for pattern recognition and
these features are Mean Absolute Value, Variance, Kurtosis, Skewness, Slope Sign Change,
Waveform length, Zero Crossing, Mean Power Spectrum Density, Median Power spectrum
Density and Root Mean Square value (RMS). Fig. 4 depicts the raw EMG signal with cor-
responding IMF’s from flexor digitorum muscle during lateral grasp. The feature selection
framework (Khushaba R. et al. 2008, Storn R. 2008, Ahmed Al-Ani et al. 2013) employed in
this work is depicted in Fig. 5.
According to Christos Sapsanis et al. (Christos Sapsanis et al. 2013) incorporation of
features derived from first three IMF’s improves the overall recognition rate. However in
our study it is found that no significant contributions in terms of classification accuracy
is received from feature set derived from IMF’s beyond second decomposition level. Some-
times feature sets derived from higher order IMF’s seems to deteriorate the overall per-
formance hence ensemble of aforementioned features from raw signal and from the first
two IMF’s taken, this feature set is denoted as TDEMD. The proposed TDEMD method
is further validated using Discrete Wavelet Packet Transform (DWPT) utilizing energy of
wavelet coefficient at each node using Daubechies family of wavelets at 4 level of decom-
position, ensemble of features extracted from statistical and auto regressive modeling (TA)
and Discrete Wavelet Transform (DWT) utilizing Standard Deviation, Entropy, Waveform
Length and Energy of wavelet coefficients using Daubechies family of wavelets at 4 level of
decomposition.
Figure 4. EMG signal with corresponding IMF’s from flexor digitorum muscle during lateral grasp.
690

Figure 5. Feature selection algorithm.
6 PATTERN RECOGNITION
There are several schemes based on artificial intelligence, statistical methods available for pat-
tern recognition. All these schemes are tested with mixed success. In order to assess the effects
of different pattern recognition algorithms on the performance of the proposed TDEMD
approach with different pattern recognition algorithms were tested in our study. Owing to
the wide acceptance in EMG based applications classifiers including Quadratic Discrimi-
nant Analysis (QDA) (Oskoei M.A. & Hu H. 2007), Support Vector Machine (LIBSVM)
(Oskoei M.A. & Hu H. 2008), K-Nearest Neighbor (KNN, k = 1) (Cover T.M. & Hart P.E.
1967), Extreme Learning Machines (ELM) (Huang et al. 2012) were utilised in this study.
The parameters of SVM are estimated by conducting grid search with cross validation.
Even though the feature selection algorithm is designed to select the optimum number of fea-
tures an ‘inner’ and ‘outer’ loop validation scheme is used to create a generalized result since the
algorithm used for feature selection is of wrapper feature selection which has dependency on the
training and testing samples. The outer loop performs the feature selection with randomly parti-
tioning the feature set into training and testing samples, 70% samples from feature set is chosen
as training samples and remaining as testing samples. This is repeated ten times and during each
iteration of the outer loop, the inner loop performs classification with selected feature subset
with 10-fold cross validation. The average results obtained are taken as the final result. This inner
and outer loop validation also helps to reduce the effect of initial population on the final result.
7.1 Comparison of different classifiers

Fig. 6 represents the average error rate when the classifiers are trained using TDEMD based
approach. The aggregated confusion matrix depicting average recognition rate of each grasp are
shown in Appendix. The misclassification rate of each grasp (average of eleven subjects) using
KNN, ELM, LIBSVM and LDA were found to be 4.79%, 2.89%, 2.08%, 5.07% respectively.
However, there are different grasps that are difficult to differentiate between each other, for
example confusions arise between S-C-H and P-L-T using KNN/LDA. But these confusions
691

Figure 6. Performance of TDEMD approach with different classifiers.
Figure 7. Performance of different feature extraction methods.
are considerably reduced using LIBSVM/ELM classifier. Among the four classifiers LIBSVM
is chosen to compare the performance of proposed TDEMD approach with other feature
extraction techniques. Though the overall results are promising it is noted that there is a subject
dependency in the recognition of each grasp which has to be further evaluated.
7.2 Comparison with other feature extraction methods

The classification error rate across all subjects using TDEMD, DWT, TA and DWPT are
shown in Fig. 7. The aggregated classification errors were found to be 2.08%, 10.1%, 14.6%,
and 19.8% for each of the TDEMD, DWPT, DWT and TA features respectively. From Fig
5, it is evident that in all subjects TDEMD approach outperformed all other feature extrac-
tion techniques. DWT, TA and DWPT provide mixed response but DWPT seems to perform
better compared to DWT and TA methods. Relatively high error rate in contrast with other
subjects exhibited by subject 5 is ascribed to issues during data acquisition.
8 CONCLUSIONS
This paper presents a preliminary report made on behalf of an ongoing research to develop
a dexterous and natural control of powered upper limbs using EMG signals. The outcome of
this stage of our research proved that the methodology of using EMD with DEFS can achieve
significantly good results. The variation of grasp recognition rate among different subjects
uncovers the requirement of fine tuning of the algorithm. Detailed analysis will be further
carried out in future with a database created by involving more subjects including amputees.
692

REFERENCES
Al-Ani, Ahmed. Alsukke, Akram., Khushaba, R. 2013. Feature subset selection using differential evo-
lution and a wheel based search strategy Swarm and Evolutionary Computation Volume 9, Pages
15–26.
Andrade, A.O., Kyberd, P.J., Nasuto, S. 2004. Time–frequency analysis of surface electromyographic
signals via Hilbert spectrum, in: S.H. Roy, P. Bonato, J. Meyer (Eds.), XVth ISEK Congress—An
Invitation to Innovation, Boston, MA, USA.
Boser, B.E., Guyon, I.M., and Vapnik, V.N. 1992. A training algorithm for optimal margin classifiers.5th
Annual ACM Workshop on COLT, Pittsburgh.
Cover, T.M., and Hart, P.E. Hart, 1967. Nearest neighbor pattern classification. IEEE Trans. Inform.
Theory, vol. IT-13, pp. 21–27.
De Luca, C.J. May 1998. The Use of Surface Electromyography in Biomechanics Journal of Applied
Biomechanics, Volume 13 Issue 2.
Ericka Janet Rechy-Ramirez and Huosheng Hu, Stages for Developing Control Systems using EMG
and EEG Signals: A survey, TECHNICAL REPORT: CES-513, ISSN 1744–8050.
Flandrin, P., Rilling, G. and Goncalv, P. 2004. Empirical mode decomposition as a filter bank. IEEE
Signal Process.Lett.vol. 11, no. 2, pp. 112–114.
Fougner, A., Stavdahl, O., Kyberd, P.J., Losier, Y.G., and Parker, P.A. Sep.2012. Control of upper limb
prostheses: Terminology and proportional myoelectric control—A review. IEEE Trans. Neural Syst.
Rehabil. Eng., vol. 20, no. 5, pp. 663–677.
Frank, F.H., 1989. Atlas of Orthopedic Anatomy. Ciba—Geigy, Switzerland.
Huang, G.-B., Zhou, H., Ding, X., & Zhang, R. 2012. Extreme learning machine for regression and
multiclass classification.IEEE Transactions on Systems, Man, and Cybernetics, Part B: Cybernetics,
42(2), 513–529.
Huang, N.E., Shen, Z., Long, S.R., Wu, M.C., Shih et al.; March 1998. The Empirical Mode Decom-
position and the Hilbert Spectrum for Nonlinear and Non stationary Time Series Analysis. Royal
Society Proceedings on Math, Physical, and Engineering Sciences, Vol. 454, No. 1971 pp. 903–995.
Jiang, N., Dosen, Muller, K.R. and Farina, D. Sep. 2012. Myoelectric control of artificial limbs—Is
there a need to change focus?[In the spotlight] IEEE Signal Processing. vol. 29, no. 5, pp. 150–152.
Khushaba, R., Al-Ani, A., Al-Jumaily, 2008. A. Differential evolution based feature subset selection.
Proceedings of the International Conference Pattern Recognition (ICPR’08).
Kilbreath, S.L., Gorman, R.B., Raymond, J. and Gandevia, S.J. Distribution of the forces produced by
motor unit activity in the human flexor digitoriumprofundus 2002. Journal of Physiology, vol. 543,
no. 1, pp. 289–296.
Kunju, Nissan., Ojha, Rajdeep. R, Suresh. Devasahayam. 2013. A palmar pressure sensor for measure-
ment of upper limb weight bearing by the hands during transfers by paraplegics. Journal of Medical
Engineering and Technology, Vol. 37, No. 7, Pages 424–428.
Kunju, Nissan. Tharion, George. R, Suresh. Devasahayam, M. Manivannan. 2013. Muscle Activa-
tion Pattern and Weight Bearing of Limbs during Wheelchair Transfers in Normal Individuals—a
step towards Lower Limb FES Assisted Transfer for Paraplegics. Converging Clinical and Engi-
neering Research on NeuroRehabilitation, pp. 197–201. Biosystems and Biorobotics Series Springer
(doi:10.1007/978-3-642-34546-3_31).
Micera, S.and Navarro, X. Jan. 2009. Bidirectional interfaces with the peripheral nervous system. Int.
Review of Neurobiology. vol. 86, pp. 23–38.
Oskoei, M. A., & Hu, H. 2007. Myoelectric control systems—a survey. Biomedical Signal Processing and
Control, 2(4), 275–294.
Oskoei, M. A., & Hu, H. 2008. Support vector machine-based classification scheme for myoelectric
control applied to upper limb. IEEE Transactions on Biomedical Engineering, 55(8).
Sapsanis, Christos, Georgoulas, Georgoulas and Tzes. 2013. Anthony Tzes, EMG based classification
of basic hand movements based on time-frequency features. 21 st Mediterranean Conference on Con-
trol and Automation (MED).
Scheme, E. and Englehart, K.. 2011. Electromyogram pattern recognition for control of powered upper-
limb prostheses: State of the art and challenges for clinical use. Journal of Rehabilitation Research
Development.vol. 48, no. 6, p. 643.
Schlesinger, G. 1919. The mechanical construction of the artificial limb Verlag von Julius Springer, pp. 321–661.
Shrirao, N.A., Reddy, N.P. and Kosuri, D.R. 2009. Neural network committees for finger joint angle
estimation from surface Emg signals. Journal of Bio Medical OnLine, vol. 15, no. 2, pp. 529–535.
Storn, R. 2008. Differential evolution research—trends and open questions, in: U.K. Chakraborty (Ed.),
Advances in Differential Evolution SCI, vol. 143, Springer-Verlag, Berlin, Heidelberg, pp. 1–31.
693

APPENDIX
The aggregated confusion matrix (Average of eleven subjects) illustrating the performance of
TDEMD method with different classifiers are attached below.
Table 1. Aggregated confusion matrix (KNN classifier).
PREDICTED
S C L T P H
S 5728 68 32 0 22 56
T C 48 5508 36 0 33 22
R L 22 40 5570 106 76 44
U T 28 24 84 5696 56 18
E P 2 12 64 96 5578 28
H 88 48 32 14 44 5792
Table 2. Aggregated confusion matrix (LIBSVM classifier).
PREDICTED
S C L T P H
S 5432 30 0 0 7 19

T C 40 5419 0 0 24 9
R L 2 0 5586 52 46 13
U T 16 0 86 5715 31 37
E P 2 8 14 26 5644 4
H 0 4 22 0 0 5526
Table 3. Aggregated confusion matrix (ELM classifier).
PREDICTED
S C L T P H
S 7944 45 0 0 9 204

T C 36 8292 0 24 24 102
R L 9 0 7992 48 108 78
U T 0 9 48 7572 96 15
E P 0 12 24 69 7188 36
H 63 33 48 60 0 7464
Table 4. Aggregated confusion matrix (LDA classifier).
PREDICTED
S C L T P H
S 5236 176 18 0 44 182

T C 126 5576 0 4 12 84
R L 62 14 5432 196 224 12
U T 16 0 156 5224 224 6
E P 32 6 52 40 5684 2
H 36 10 2 2 8 5620
694

Automated aquaponics system
B. Sreelekshmi & K.N. Madhusoodanan

Department of Instrumentation, Cochin University of Science and Technology, Cochin, India
ABSTRACT: Agriculture has proven to be an arena in which technology has crucial roles
to play. Presently, agriculture automation has a wider scope with emerging trends such as
Controlled Environment Agriculture, Precision Farming etc. This paper describes the design
and implementation of an automated aquaponics system in the framework of Internet of
Things. Aquaponics is the integration of recirculating aquaculture with hydroponics. It is a
sensitive system in which several parameters are to be maintained at certain optimum values
inorder to ensure proper functioning of the system. The implemented automated aquaponics
system enables web-based remote monitoring of the important parameters via the ThingSpeak
IoT platform using Arduino Uno, ESP8266-01 and several sensors. Arduino-based control
of certain parameters is also possible in case of deviation from setpoint. In addition, there is
an SD-card data-logger used to store the data for future analysis.
1 INTRODUCTION
Integration of technology with agriculture has remarkably increased the ease and efficiency of
agriculture. Presently, technology-related agriculture is adopting newer dimensions such as Con-
trolled Environment Agriculture, Precision Farming (Mondal & Basu 2009) etc. These emerging
agricultural trends make use of various technological aspects such as WSN (Ojha et al. 2015),
IoT, Artificial Intelligence (Hashimoto et al. 2001, Lee 2000), control systems (De Baerdemaeker
et al. 2001) and so on with an aim to improve factors such as sustainability and food security.
Aquaponics is the integration of recirculating aquaculture with hydroponics in a single sys-
tem (Diver 2000). It is a form of Controlled Environment Agriculture and is thus a technology-
related agricultural practice. Aquaculture is the rearing of fish in controlled conditions whereas
hydroponics involves soilless growth of plants. Thus, the combination of both the techniques
into a single system enables the organisms to benefit mutually wherein the plants absorb the
required nutrients and the fishes are provided with purified water. The aquaculture effluent
consists of ammonia which is toxic to fish. The water from the aquaculture tank is pumped
to a grow-bed (which serves as the substrate for plant growth) and recirculated back into the
aquaculture system with the help of a siphon. During the circulation, the water is subjected to
a two-step nitrification process (Somerville et al. 2014) in which Nitrosomonas bacteria converts
ammonia into nitrite which is then converted into nitrate, an absorbable nutrient for plants, by
Nitrobacter bacteria (Klinger & Naylor 2012). The nitrate is absorbed by the plants (Buzby &
Lin. 2014) and the filtered water is recirculated back into the aquaculture tank (Graber & Junge
2009, van Rijn 2013) with the help of a flood and drain mechanism operated by a siphon.
For the aquaponics system to be properly balanced, several parameters should be maintained
at certain optimum values. The important parameters include temperature, humidity, light
intensity, water level in the aquaculture tank, pH, nitrate and ammonia content, dissolved
oxygen level etc. Regular manual monitoring and control of such parameters is a difficult
task for the farmers (Goddek et al. 2015). The requirement of an automated aquaponics
system (Saaid et al. 2013) lies in this aspect.
This paper describes the design and implementation of an automated aquaponics system
in the framework of Internet of Things. The implemented system enables remote monitoring
695

of the important parameters via the ThingSpeak IoT platform. Two wireless sensor nodes are
deployed in the aquaponics system (one in the aquaculture tank and other in the growbed).
Each sensor node consists of an Arduino Uno, several sensors and an ESP8266-01 Wi-Fi
transceiver. The Wireless Sensor Network reports the sensor readings to the ThingSpeak IoT
platform via the Wi-Fi for real-time monitoring purpose. An arduino-based control of water
level in the aquaculture tank is also provided to maintain optimum water level. In addition,
there is an SD-card data-logger for storing the sensor data into a microSD card for future
analysis. The farmers can thus perform an analysis of the important parameters at different
stages so as to seek better farming strategies.
2 OVERALL SYSTEM ARCHITECTURE
2.1 Small-scale IBC aquaponics system

A conventional small-scale IBC (Intermediate Bulk Container) type aquaponics system was
designed and implemented inorder to study the various aspects of the system. The dimen-
sions (shown in Figure 1) and design features of the implemented system were chosen as
described by Somerville et al. (2014).
The design technique adopted was Media-Bed technique which is the most popular
aquaponic design technique (Lennard & Leonard 2006). A medium like clay, metal etc. serves
as substrate for plant growth. The fish species used was GIF (Genetically Improved Farm-
ing) Tilapia. The plants used Solanum lycopersicum (tomato), Phaseolus vulgaris (beans) and
Abelmoschus esculentus (Ladies finger). A bell siphon was used for recirculating the water
from the media-bed into the aquaculture tank through a flood and drain mechanism.
2.2 Architecture of automated system

The implemented automated system includes three subsystems deployed in the IBC Aquapon-
ics farm: an IoT-based Aquaponics Monitoring System, Arduinobased level contol and an
SD-card data-logger.
2.2.1 IoT-based aquaponics monitoring system

This part enables real-time monitoring of the important parameters via the ThingSpeak IoT
platform. Two sensor nodes are deployed in the IBC aquaponics system (one in the grow bed
and other in the aquaculture tank). Each sensor node consists of an Arduino Uno (micro-
controller), several sensors and ESP8266-01 Wi-Fi transceiver. The Wireless Sensor Network
is responsible for reporting the sensor readings to the ThingSpeak IoT platform via the Wi-Fi
for realtime monitoring purpose. The parameters monitored include ambient light intensity,
ambient temperature, relative humidity, grow-bed moisture, level and temperature of water
in the aquaculture tank.
Figure 1. Dimensions of IBC aquaponics system (Top view).
696

Figure 2. Arduino-based level control.
Figure 3. Overall system architecture of automated aquaponics system: IoT-based monitoring system
and Arduino-based control.
2.2.2 Arduino-based level control

The level of water in the aquaculture tank is a crucial parameter while considering an
aquaponics system. An outdoor aquaponics system which is prone to variant environmental
conditions demands an accurate level control mechanism inorder to maintain optimum water
level. The proposed level control system (as shown in Figure 2) employs Arduino Uno as the
microcontroller and a two channel 5 V relay module is used to give the actuating signal to a
pump depending upon the deviation of water level from the set-point value (either increase or
decrease).
2.2.3 SD-card data-logger

The sensor readings can be stored in a micro-SD card by deploying an SD card data-logger
in the aquaponics system. An SD-card module (which holds the micro-SD card) is interfaced
with Arduino Uno. The sensor data stored in micro-SD card can later be imported as excel
files for future analysis.
The overall conceptual diagram of the implemented automated aquaponics system is
shown in Figure 3.
3 SYSTEM DESIGN
The hardware and software design of the implemented system is briefly described in this
section.
697

3.1 Hardware design
The hardware requirements for the real-time monitoring purpose include Arduino Uno,
sensors and ESP8266-01 Wi-Fi transceiver module. The Arduinobased level controller
requires a two channel 5 V relay module as the actuator. In addition, an SD-card module is
required for the SD-card data-logger.
3.1.1 Arduino Uno

Arduino Uno is a microcontroller board based on ATmega328P. It consists of 16 MHz
quartz crystal, USB connection, powerjack, an ICSP header and reset button. It has 14
digital input/output pins (out of which six can be used as PWM outputs) and six analog
input pins. Operating voltage is 5 V.
3.1.2 Sensors
Six parameters of the aquaponics system are monitored, namely ambient light intensity,
ambient temperature, relative humidity, moisture content in the grow-bed, level and tempera-
ture of water in the aquaculture tank. Five different sensors are used for this purpose. LDR
(Light Dependent Resistor) is used to measure the ambient light intensity. DHT11 is the
sensor used for measuring the ambient temperature (in degree celsius) and relative humidity
(in percentage). The grow-bed moisture sensor consists of two probes which act as a variable
resistor depending upon the moisture content of the grow-bed media. Ultrasonic level sensor
HCSR04 is used for measuring the level of water in the aquaculture tank. The temperature of
water in the aquaculture tank is measured using DS18B20 waterproof temperature sensor.
3.1.3 ESP8266-01 Wi-Fi module

ESP8266-01 is a Wi-Fi transceiver module widely used for IoT applications. This module
enables Wi-Fi access to microcontrollers. The operating voltage is 3.3 V. In the proposed
system, ESP8266-01 is used to send sensor data into ThingSpeak IoT platform.
3.1.4 Two channel 5 V relay module

The 5 V relay module can be controlled directly by the Arduino microcontroller. For a relay
to be switched ON, the digital output given from arduino should be LOW. A HIGH digital
output from arduino will switch the relay module to OFF position.
3.1.5 SD card module

The SD card module is interfaced to Arduino Uno to store the sensor data. A micro-SD card
is mounted on the SD card module for storing data. The operating voltage is 5 V.
3.2 Software requirements

The software requirements of the system include Arduino IDE and ThingSpeak.
3.2.1 Arduino IDE

Arduino Integrated Development Environment (IDE) is an open-source software used to
upload programs into the arduino hardware and communicate with them. The programs
written in Arduino IDE are called sketches. These sketches are written in the text editor and
are saved with the file extension ‘.ino’. The output is displayed in the serial monitor.
3.2.2 ThingSpeak
ThingSpeak is an open-source IoT platform used for real-time monitoring purposes. ThingSpeak
channels can be created by logging into ThingSpeak using MathWorks account. The data
from the sensors get stored in ThingSpeak channels in various fields (Pasha 2016, Rao & Ome
2016). The data is displayed in the form of charts. Certain details are required to be entered
in the Arduino IDE sketch inorder to send sensor data from Arduino to ThingSpeak using
ESP8266-01 Wi-Fi module. These include the write API key of ThingSpeak channel, Thing-
Speak IP, the SSID and password of the Wi-Fi network to be accessed and HTTP GET request.
698

Real-time monitoring of the parameters of the automated system is possible from any remote
location by logging into the ThingSpeak channel created using Mathworks account. The link to
the ThingSpeak login page is https://thingspeak.com/login. ThingSpeak channel consists of fields
corresponding to each parameter. The measured readings are displayed in the form of charts.
Figure 4 shows the real-time graphical display of measured parameters as obtained
from the ThingSpeak channels. The y-axis of all the charts are labelled with the parameter
concerned and the x-axis is labelled as date with the time at which reading is taken.
The status of actuation of the arduino-based level controller can also be monitored
using ThingSpeak channel. Two fields are created corresponding to each relay of the two
channel relay module. When a relay turns ON, the chart corresponding to the particular relay
displays 1. When the relay turns OFF, the status displayed in the chart changes to zero.
Figure 4. Fields of ThingSpeak channel showing the real-time values of measured parameters against
corresponding time.
Figure 5. Plot of measured parameters aginst time from data stored in microSD card using SD-card
datalogger.
699

The data stored in the micro-SD card using the SD card data-logger is imported and saved
as excel files for future analysis purpose. A plot of the measured parameters against time for
a period of twenty-four hours is shown in the Figure 5.
5 FUTURE DIRECTIONS AND CONCLUSIONS
The proposed automated system can be modified by including other sensors such as dissolved
oxygen, pH and ammonia sensors. The same methodology can be adopted to other Controlled
Environment Agriculture practices such as Greenhouse, Hydroponics and Aquaculture. Auto-
mation can also be carried out using Wireless Sensor Networks employing Raspberry Pi and
zigbee protocol (Ferdoushi & Li 2014) wherein image processing techniques can be applied for
analysis of effects of various environmental factors on plant growth (Liao et al. 2017). Image
processing techniques can also be applied for weed detection. Intelligent controllers utilizing
Artificial Neural Networks and Expert Systems can also improve the performance of Control-
led Environment Agriculture systems to a greater extent (Hashimoto et al. 2001, Lee 2000).
The scope of technology related agricultural practices is increasing due to their sustainable
nature and efficient utilization of resources such as land, water etc. Application of technological
aspects enable the farmers to adopt better farming strategies and thereby increase the food pro-
ductivity. These methods are thereby capable of ensuring sustainability as well as food security.
REFERENCES
Buzby, K. & L.-S. Lin. (2014). Scaling aquaponic systems: Balancing plant uptake with fish output.
Aquacultural Engineering 63, 39–44.
De Baerdemaeker, J., A. Munack, H. Ramon, & H. Speckmann (2001). Mechatronic systems, commu-
nication, and control in precision agriculture. IEEE Control Systems Magazine, 48–70.
Diver, S. (2000). Aquaponics integration of hydroponics with aquaculture. Technical report, ATTRA, NCAT.
Ferdoushi, X. & X. Li (2014). Wireless sensor network system design using raspberry pi and arduino for
environmental monitoring applications. Procedia Computer Science 34, 103–110.
Goddek, S., B. Delaide, U. Mankasingh, K. Ragnarsdottir, H. Jijakli, & R. Thorarinsdottir (2015).
Challenges of sustainable and commercial aquaponics. Sustainability 7, 4199–4224.
Graber, A. & R. Junge (2009). Aquaponic systems: Nutrient recycling from fish wastewater by vegetable
production. Desalination 246, 147–156.
Hashimoto, Y., H. Murase, T. Morimoto, & T. Torii (2001). Intelligent systems for agriculture in japan.
IEEE Control Systems Magazine, 71–85.
Klinger, D. & R. Naylor (2012). Searching for solutions in aquaculture: Charting a sustainable course.
Annual Review of Environment and Resources 37, 247–276.
Lee, P.G. (2000). Process control and artificial intelligence software for aquaculture. Aquacultural Engi-
neering 23, 13–36.
Lennard, W. & B. Leonard (2006). A comparison of three different hydroponic sub-systems (gravel bed,
floating and nutrient film technique) in an aquaponic test system. Aquacult. Int. 14, 539–550.
Liao, M., S. Chen, C. Chou, H. Chen, S. Yeh, Y. Chang, & J. Jiang (2017). On precisely relating the
growth of phalaenopsis leaves to greenhouse environmental factors by using an iot-based monitoring
system. Computers and Electronics in Agriculture 136, 125–139.
Mondal, P. & M. Basu (2009). Adoption of precision agriculture technologies in India and in some
developing countries: Scope, present status and strategies. Progress in Natural Science 19, 659–666.
Ojha, T., S. Misra, & N.S. Raghuwanshi (2015). Wireless sensor networks for agriculture: The state-of-
the-art in practice and future challenges. Computers and Electronics in Agriculture 118, 66–84.
Pasha, S. (2016). Thingspeak based sensing and monitoring system for iot with matlab analysis. IJNTR
2, 19–23.
Rao, S. & N. Ome (2016). Internet of things based weather monitoring system. IJARCCE 5, 312–319.
Saaid, M., N. Fadhil, M. Ali, & M. Noor (2013). Automated indoor aquaponic cultivation technique.
Proc. of 2013 IEEE 3rd International Conference on System Engineering and Technology., 285–289.
Somerville, C., M. Cohen, E. Pantanella, A. Stankus, & A. Lovatelli (2014). Small-scale aquaponics food
production. Technical report, Food and Agricultural Organization of the United Nations, Rome.
van Rijn, J. (2013). Waste treatment in recirculating aquaculture systems. Aquacultural Engineering 53, 49–56.
700

A review of the methods for despeckling in optical

coherence tomography
K. Athira, K. Brijmohan, Varun P. Gopi, K.K. Riyas, Garnet Wilson & T. Swetha
Department of Electronics and Communication Engineering, Government Engineering College, Wayanad, India
ABSTRACT: In medical image processing, the Optical Coherence Tomography (OCT)

imaging technique is widely used for disease detection. It is a non-invasive imaging technique
that provides images of tissue structures with high resolution. Speckle noise is a granular
noise. Speckle noise is introduced in OCT images due to the constructive and destructive
interference of optical waves that undergo multiple scattering in different directions, when
waves propagate through the tissue. The quality of OCT images degrades due to speckle
noise. This paper is a review of different methods for speckle noise reduction in OCT images.
Keywords: Speckle Noise; Optical coherence Tomography; Denoising Filters
1 INTRODUCTION
Optical coherence tomography (OCT) is a non-invasive imaging technique that provides high
resolution images of tissue structures and cross-sectional imaging of many biological systems
(Drexler & Fujimoto, 2008). The OCT imaging technique is widely used by ophthalmologists
in the diagnosis of eye disease such as Glaucoma, Macular Edema and Diabetic Retinopathy.
Now-a-days this imaging technique is also used in the detection of skin disorders.
The working of the OCT imaging technique is based on the Michelson Interferometer,
using Low Coherence Interferometry (Schmitt et al., 1999). Typically, near-infrared laser
light is used as a light source to penetrate into the scattering medium, before capturing
the backscattered optical waves. Due to heat produced by the image sensors, or due to
the physical properties of light photons the image is corrupted during the image acquisi-
tion process. The light reflected from the micro-structural tissue contains the features of
the image. The combination of various crests and troughs of backscattered light waves
from the tissue produces granular structures in an image. This grainy representation is
known as speckle noise. It can change the important details in an image used to diagnose
disease. Therefore leads to image quality degradation. This degradation makes it difficult
for humans to differentiate pathological tissues from the normal tissues. Speckle noise is a
multiplicative noise which contains information about the image. Therefore it is difficult to
remove speckle noise without any change in important features in an image. The primary
aim of the OCT research is to denoise the speckle noise and preserve the edges clearly.
Several filtering methods are proposed for reducing speckle noise. The limitation is that
filtering techniques remove some parts of the information in an image along with speckle
noise. This paper presents a comparative study on the performance of different filters.
2 NOISE MODEL
Speckle noise can be modeled as multiplicative noise. It is known to have a Gamma distribu-
tion. It is a granular type noise which appears in the lighter regions of the image as bright
specks. Speckle noise can be modeled as:
701

Y(x, y) = S(x, y).N(x, y)
where Y, S and N represent the noisy image, signal and speckle noise, respectively. A logarith-
mic transformation is applied to the image data to change the multiplicative nature. Then the
model can be rewritten as:
f(x,y) = s(x,y) + e(x,y)
where f, s and e represent the logarithm of the noisy image, signal and noise respectively.
3 RELATED WORKS
3.1 Nonlocal means denoising filter with double Gaussian anisotropic kernels
The Non-local means (NLM) filter is one of the important denoising filters (Aum et al., 2015).
It is a denoising algorithm which utilizes the presence of similar features in an image and then
takes the average of those features to remove speckle noise in an OCT image. This method
provides a low signal-to-noise ratio due to the low performance of noise reduction around the
edges of an image. To overcome this limitation, the conventional NLM filter converts into an
NLM filter with double Gaussian anisotropic kernels. The conventional NLM filter contains
a Gaussian kernel to measure the similarities in an image. It may be able to measure the dis-
tinct similar features from the image. Since the same Gaussian kernel is used on every pixel,
the speckle noise corrupted edges cannot be denoised correctly. Therefore, the new algorithm
proposes new kernels and their shapes are adaptively varied. Various kernels were used for cal-
culating the similarity between the local neighborhoods from the pixel positions. The modified
NLM method produced a PSNR of 31.01db. Figures 1(a) and 1(b) show the denoised images,
using the conventional NLM filter and the modified NLM filter, respectively.
3.2 Noise adaptive wavelet thresholding

The working of optical the OCT imaging technique is based on coherence detection of inter-
ferometric signals. Speckle noise is introduced into the OCT image, due to the constructive
and destructive interference of optical waves that undergo multiple scattering in different
directions. Wavelet domain thresholding provides much noise suppression by preserving
image sharpness whilst removing speckle noise in the OCT image. Speckle noise has dif-
ferent characteristics, but it has not been considered in conventional wavelet thresholding
algorithms. A Noise Adaptive Wavelet Thresholding (NAWT) algorithm is introduced in this
paper (Adler et al., 2004). Graphical representation of the optimized adaptive wavelet thresh-
olding algorithm is shown in the Figure below.
The algorithm uses two types of images:- the reference image and the original OCT image.
The speckle noises can be characterized by using the reference image received from a uniform
scattering sample, since the variation in magnitude for such an image attributes to random
noise. Wavelet transform is applied to both the original image and the reference image. Then
the signal variance and noise variance are calculated in all subbands:-
Figure 1. OCT images obtained from a human index fingertip: (a) image processed with the conven-
tional NLM: - (b) image processed with the modified NLM.
702

T = σ w2 / σ x
where T is the thresholding value. Thresholding is applied on an OCT image and performing
inverse wavelet transform.
3.3 Wavelet denoising of multiframe OCT image

In this method, the wavelet technique is used for denoising of multiframe OCT data (Mayer
et al., 2012). This algorithm makes use of wavelet decompositions in single frames for noise
estimation, instead of taking the average of multiple image frames. The flow diagram of a
multiframe denoising algorithm is shown below.
Speckle noise is a multiplicative noise. The preprocessing stage is logarithmic transforma-
tion, which is applied to the input image frames. The multiplicative noise is therefore con-
verted into additive noise. The single frames are decomposed by wavelet transformations.
Two different wavelet transformations are used and compared with each other such as with
discrete stationary wavelet transformation and dual tree complex wavelet transformation.
A significance weight and a correlation weight are proposed in this algorithm. The sig-
nificance weight provides local noise estimation and correlation weight gives details of a
structure. Finally, the despeckled image is attained by taking the inverse wavelet transform.
3.4 Combination of wiener filter and wavelet transform

This method proposes an effective algorithm based on the combination of spatial and fre-
quency domain techniques for despeckling in OCT image (Rajesh et al., 2016). The wiener
filtering method is applied to the speckled image as a preprocessing stage. This filter works
efficiently for the suppression of additive and multiplicative noise in an image. The speckle
Figure. 2. Flow chart of the adaptive thresholding algorithm.
Figure 3. Graphical representation of the wavelet multiframe denoising algorithm.
703

noise model is multiplicative in nature. The resultant image of the wiener filtering is trans-
formed into an additive one. Then discrete wavelet transform is applied to this output image.
Thresholds are computed for each sub band except for the lowest level. Then thresholding
is applied to all sub bands and inverse discrete wavelet transform is applied. The exponent is
taken to get the despeckled image.
3.5 Adaptive bilateral filtering technique

The bilateral filter is proposed by Tomasi and Manduchi [7]. This non-linear filtering
technique smoothes the edges as well as preserving the edge structures. The kernel of the
conventional bilateral (CB) filter contains two components, such as a range filter kernel and
a domain filter kernel. CB filter response at a given pixel location is given as:
1
Iˆ ( m ) = ∑ wD ( m, n) wR ( m, n) I ( n),
Zn ∈N (m)
The weight function WD is related to the domain filter which provides larger weight to
pixels that are spatially close to the center pixel. Similarly, the other weight function WR
Table 1. Comparison of despeckling algorithm.
Algorithm Properties Remarks
Non-local Estimates pixel intensity It is not able to suppress speckle noise

mean based on the presence for non—repetitive neighborhoods.
of similar patterns and It does not preserve edge details.
features of an image.
Modified Using non-local means It is able to suppress speckle noise
Non-Local denoising filter with at the edges of an image.
mean double Gaussian
anisotropic kernels.
The shape of Gaussian
kernel varies
adaptively.
Noise adaptive Based on variance It eliminates a significant amount
wavelet of reference image of noise.
Thresholding and OCT image the It does not preserve edge details
thresholding value completely.
is estimated in all Signal-to-Noise ratio is less.
subbands.
Wavelet denoising This algorithm makes It removes the noise but accuracy
of multiframe use of wavelet depends on number of input frames.
OCT decompositions in
single frames for noise
estimation instead of
taking the average of
multiple image frames.
Combination of Algorithm based on It is able to suppress both additive
wiener filter and the combination of and multiplicative noise in an image.
wavelet transform spatial and frequency
domain techniques for
reducing speckle noise
Adaptive Bilateral Combination of two Eliminates significant amount of
Filtering filters based on noise.
Technique spatial distance Performs better in preserving sharp
and intensity edges and fine details.
difference
704

is linked to range filtering and the range lowpass Gaussian filter provides larger weight to
pixels. These pixels are the same as the center pixel in gray value. These weight functions can
be defined as:
 m−n2
wD ( m, n) = exp  − 
 2σ d2 
 I −I 2
wR ( m, n) = exp  − m 2 n 
 2σ r 
where Im and In are the intensities at m and n, respectively. The σd is the geometrical spread
in the domain which calculates the blurring effect in an image. Then the value of σd is inde-
pendent to noise. Similarly, the σr is the photometric spread in the image range and its opti-
mal value is linearly proportional to the noise standard deviation σ. The photometric spread
in the range filter is a constant value. This is one of the main disadvantages of the bilateral
filter, as fixing the optimal value is difficult. In order to enhance the sharpness of an image,
the bilateral filter needs some modifications. It contains two modifications:- an offset is intro-
duced to the range filter, and the width of the range filter varies in an adaptive manner.
4 CONCLUSION
The speckle noise reduction algorithms are described in this paper. Among filters used for
despeckling in optical coherence tomography images, the bilateral filtering technique is
efficient. This technique eliminates a significant amount of noise and preserves the edges of
the denoised image. Adaptive filter methods are more efficient than filtering as they preserve
the fine details of edges.
REFERENCES
[1] W. Drexler, J.G. Fujimoto, “Optical Coherence Tomography: Technology and Applications”,
Springer International Publishing, Switzerland, 2015.
[2] J.M. Schmitt, S.H. Xiang, and K.M. Yung, “Speckle in Optical coherence Tomography: An Over-
view” J. Biomed. Opt. 4 (1) (1999) 95–105.
[3] Jaehong aum, Ji-hyun kim, and Jichai jeong, “Effective speckle noise suppression in optical coher-
ence tomography images using nonlocal means denoising filter with double Gaussian anisotropic
kernels” published 3 April.
[4] Desmond C. Adler, Tony H. Ko, and James G. Fujimoto “Speckle reduction in optical coherence
tomography images by use of a spatially adaptive wavelet filter”, December 15, 2004/Vol. 29,
No. 24/OPTICS LETTERS.
[5] Markus A. Mayer, Anja Borsdorf, Martin Wagner, Joachim Hornegger, Christian Y. Mardin, and
Ralf P. Tornow, “Wavelet denoising of multiframe optical coherence tomography data”, 1 March
2012/Vol. 3, No. 3/BIOMEDICAL OPTICS EXPRESS.
[6] Rajesh Mohan R., S. Mridula, P. Mohanan, “Speckle Noise Reduction in Images using Wiener
Filtering and Adaptive Wavelet Thresholding”, 978-1-5090-2597-8/16/$31.00_c 2016 IEEE.
[7] Ch. Ravi Kumar, Member, IACSIT and S.K. Srivatsa, “Enhancement of Image Sharpness with
Bilateral and Adaptive Filter”, International Journal of Information and Education Technology,
Vol. 6, No. 1, January 2016.
[8] A. Ozcan, A. Bilenca, A.E. Desjardins, B.E. Bouma, G.J. Tearney, “Speckle reduction in optical
coherence tomography images using digital filtering,” J. Opt. Soc. Am. A 24 (7) (2007) 1901–1910.
[9] V. Frost, J. Stiles, K. Shanmugan, J. Holtzman, “A model for radar images and its application to
adaptive digital filtering of multiplicative noise,” IEEE Trans. Pattern Anal. Mach. Intell. (PAMI-4)
(2) (1982) 157–166.
[10] T. Loupas, W. McDicken, P. Allan, “An adaptive weighted median filter For speckle suppression in
medical ultrasonic images,” IEEE Trans. Circuits Syst. 36 (1) (1989) 129–135.
705

Generic object detection in image using SIFT, GIST

and SURF descriptors
Dona P. Joy, K.S. Shanthini & K.V. Priyaja

Department of Applied Electronics and Instrumentation Engineering, Government Engineering College,
Calicut, Kerala
ABSTRACT: This paper compares generic object detection method using three different
feature extraction schemes. The query image could be of different types such as a real image
or a hand-drawn sketch. The method operates using a single example of the target object.
The feature descriptors emphasizes the edge parts and their distribution structures, so it
is very robust and can deal with virtual images or hand-drawn sketches. The approach is
extended to account for large variations in rotation. Good performance is demonstrated on
several data sets, indicating that the object was successfully detected under different imaging
conditions.
Keywords: DSIFT, keypoint, SURF
1 INTRODUCTION
In image processing, it is very important to analyze the visual objects in the image. Object
detection means, to locate object of a certain class in a test image. This method uses only
one query image as the template to detect the object without any training procedures. Such
systems are applicable in different areas such as surveillance, video forensics and medical
image analysis and so on.
The training free object detection with one query image has many applications such as
automatic passport control at airports, where a single photo in the passport is the only
example available. Another application is the image retrieval from the Web. In this case, only
a single sample of the target is provided by the user and every database is compared with this
single sample. Another application is for the classification of an unknown set of images into
one of the training classes.
2 LITERATURE REVIEW
There are different methods for object detection. Most of them are based on training process.
But the training-based methods are subject to sample restrictions. In cases such as frontal
face detection, samples can accurately get aligned. But, in many cases, the collecting and
aligning of samples is not possible which badly affects the performance of training-based
detection methods. And the training method is not suitable for immediate task, because the
collection of samples and training the model should be completed in advance. So when the
target class changes, they must be redone.
2.1 Keypoint-based matching

Most methods use several key points in the image which are relatively stable and calculate
the local invariant descriptor of the patch near the key point. Under these circumstances,
707

the object image is converted into a set of local descriptors on the key points. Mikolajczyk
and Schmid [2] had compared these methods, with SIFT algorithm shown to outperform
other keypoint-based descriptors. And keypoints provide greater invariance and more com-
pact coding, but they are not designed to select the most informative area for detection [3].
Another drawback is the object detection may have multiple objects or no objects in the
given test image. So it complicates locating each object in the test image from the detected
keypoints in the case of keypoint-based methods.
2.2 Densely computed descriptors

The local self-similarity descriptor [4] can be used as a descriptor for intensive calculations. It
computes a simple sum of the squared differences between a center image patch and periph-
eral image patches. Some methods used a local regression kernel [4] as the descriptor. Rather
than a point based representation, these densely computed descriptors give a global represen-
tation of the object. But these methods are not suitable to handle with hand-drawn images,
which may differ from the real image in appearance.
2.3 Contour segment network

Contour-based methods such as the Contour Segment Network [6] can be used to detect
object using a single hand drawn sketch. In these methods the original image is converted
into a group of contour segments similar to the hand-drawn sketch. The object detection
problem is developed as finding paths through the network resembling outlines of the model.
The method allows the scale of the image to change and is computationally efficient. The
disadvantage is that the contour detection is usually not very stable with complex background.
Because, for complex background the generation of contour segments becomes very difficult.
2.4 Other descriptors

Zhang [10] proposed an efficient image matching technique based on SURF descriptor. They
compared SURF to SIFT and found that SURF was more robust compared with SIFT and
has a fast matching speed.
Most commonly developing object detection methods are based on local and global fea-
tures. Global features are characteristics of regions in images such as area, perimeter, Fourier
descriptor and moments. The local features are usually based on the boundary of an image
or represent distinguishable small area. GIST features are a type of global feature. The GIST
was first introduced by Murphy et al. [11]. They took global feature of an image which has
been called as gist of a scene to overcome the ambiguity caused by only taking the local frag-
ments of an image and interest point detector around it for object detection. In Ivan et al.
[12], they described the effectiveness of GIST features in image classification. GIST describes
the shape of the scene using low dimensional feature vectors, and it performs well in clas-
sification problems.
3 OVERVIEW
Object detection means, to locate any object of a particular class in a test image. This method
uses only one query image as the template to detect the object without any training proce-
dures. As shown in Figure 1, the query image should be a typical sample of the target class,
containing only one object and as little background as possible. It can be a real image, a vir-
tual image from a simulation model or even a hand-drawn sketch which only exhibits a rough
profile of the object. The detection task is very similar to the template matching process. The
query image is used as a standard template and the test images are matched to this template
to recognize the objects.
708

Figure 1. Object detection example using a single query image.
The test image T, is divided into overlapping patches, Ti, which have the same size as Q.
Then the features of query image and the test image patches are extracted. These query image
is compared with each of the test image patches and the most silimar patch is decided.
4 THEORY
The main steps of the object detection method are feature extraction, dimensionality reduc-
tion, similarity measurement and decision making, as shown in Figure 2. Feature extraction
is the important step in the object detection process. The proposed method use three different
feature extraction methods-Dense Scale Invariant Feature Transform (DSIFT), GIST and
Speed UP Robust Features (SURF). Principal Component Analysis (PCA) is used to reduce
the dimensionality of the features. Euclidean distance and Matrix Cosine Similarity (MCS)
are used for similarity measurement. Decision process is based on minimum Euclidean dis-
tance or maximum MCS value.
4.1 Feature extraction

Feature extraction is an important step in object detection. Features should be easy to
extract from images and unique to particular object. For the test image data, the features
are extracted and compared with the query image features. In the proposed method, SIFT,
GIST and SURF features are used for this purpose. The test image is divided into overlap-
ping patches, Ti. Then features of the query image and test image patch are compared to find
the similarity.
1. DSIFT: For the given image SIFT finds all the keypoints in the image with respect to the
gradient feature of each pixel. Every keypoint contains the information of its location,
local scale and orientation. Then, based on each keypoint, SIFT computes a local image
descriptor which shows the gradient feature in the local region around the keypoint.
Combining all the local descriptors, we get the complete features from the image. Densely
computed SIFT (DSIFT) descriptors are very accurate and fast than SIFT. And for dense
SIFT, the location of each keypoint is not from the gradient feature of the pixel, but from
a pre-designed location. The DSIFT can grasp the most general characteristics of the
object. The SIFT descriptors are calculated on every small patch of the image as shown
in Figure 3, centered at dense sampling points [1]. Then, these densely computed SIFT
features are arranged together to form the DSIFT descriptor.
2. GIST: GIST features are computational model of the recognition of scene categories that
bypasses the segmentation and the processing of objects. The GIST descriptor focuses on
the shape of scene itself, on the relationship between the outlines of the surfaces and their
properties, and ignores the local objects in the scene and their relationships. The represen-
tation of the structure of the scene, termed spatial envelope is defined, as well as its five
perceptual properties: naturalness, openness, roughness, expansion and ruggedness, which
are meaningful to human observers. The input image is first pre-processed by converting
it to grayscale, normalizing the intensities and locally scaling the contrast. The resulting
709

Figure 2. Block diagram of the prposed system.
image is then split into a grid on several scales, and the response of each cell is computed
using a series of Gabor filters. All of the cell responses are concatenated to form the fea-
ture vector.
3. SURF: SURF descriptors can be used to locate and recognize objects, people or faces, to
reconstruct 3D scenes, to track objects and to extract points of interest. SURF is a fast
feature extraction method based on the integral image and Hessian matrix and it is par-
tially inspired from SIFT. To detect interest points, SURF uses an integer approximation
of the determinant of Hessian blob detector. Its feature descriptor is based on the sum of
the Haar wavelet response around the interested points.
SURF uses multi-resolution pyramid technology to convert images into coordinates to
copy the original image with pyramid-shaped Gaussian or Laplacian pyramid shapes to
obtain an image with the same size but reduced bandwidth. Thus achieves a special blur-
ring effect on the original image, called Scale-Space and ensures that the interested points
are invariant to scale.
4.2 Dimensionality reduction

Principal Component Analysis (PCA) is one of famous techniques for dimension reduction,
feature extraction, and data visualization. In general, PCA is defined by a transformation of
a high dimensional vector space into a low dimensional space. PCA provides an efficient way
to reduce the dimensionality, so it is much easier to visualize the shape of data distribution.
4.3 Similarity measurement

The next step in the proposed system is a decision rule based on the measurement of dis-
tance between the computed features. The method used Matrix Cosine Similarity (MCS) and
Euclidean distance to find the similarity between the features.
4.4 Decision process

1. Matrix Cosine Similarity: Cosine similarity is a measure of an inner product space that
measures the cosine of the angle between them. The popularity of cosine similarity due to
the reason that it is very efficient to evaluate, especially for sparse vectors, as only the non-
zero dimensions need to be considered. Cosine similarity(ρ) between two feature vectors
FQ and FT is given by:
FQ FTi
( )
ρ = ρ FQ , FTi = ,
|| FQ || || FTi ||
(1)

710

where FQ is the feature descriptor for the query image and FT is the descriptor for the test
image patch.
The detection of the object is given by finding the most similar patch to the query image
which has the similarity equal to 1.
2. Euclidean distance: The Euclidean distance is the straight line distance between two points
in Euclidean space.
di = || FQ − FTi ||2 (2)

The patch with minimum Euclidean distance will be detected as the result.
In order to demonstrate the identification performance, the system was tested for car detec-
tion, face detection, generic object detection and rotated object detection. For rotated object
detection, generic objects were implanted in the background images.
5.1 Car detection

For the car detection tests, the UIUC car data set [13] was used. The data set includes 309
images of 100 × 40 pixels. Some detection examples are shown in Figure 3 with a real image
or a hand-drawn sketch as the query image. The results show that the method is able to detect
cars with a single query image and is robust to object variations, cluttered background, and
partial occlusions.
5.2 Face detection

The system was also tested for face detection. The data is from the CMU data set [14].
The results are shown in Figure 4. There were images with one face or more than one faces.
Hand-drawn sketch gives better results for the detection process than the real image.
Figure 3. Car detection results.
711

Figure 4. Face detection results.
Figure 5. Generic object detection results.
5.3 Generic object detection

The proposed method can also be used for more generic object detection tasks. Coil-20P
dataset [15] contains 20 object classes. Each class contains 72 gray scale images of a specific
object in different positioning. All the images have pixel value 128 × 128, in PNG format.
These object images are implanted in the background images. Coil-20P is obtained from
Columbia Object Image Library. Figure 5 shows some examples of detecting the object(cup)
with different background scenes.
5.4 Rotated object detection

For rotated object detection the Coil-20P objects with rotation are implanted on different
backgrounds. Each of the objects has 72 images with 5° of rotation. The detected results are
shown in Figure 6. When the object is rotated the detection is possible only upto some degree
of rotation. This range of detection is shown in Table 1 for 4 different objects using MCS as
the similarity measurement.
712

Figure 6. Rotated object detection results.
Table 1. Result for rotated object.
Detected range (in degree)
Object DSIFT GIST SURF
Glass 0–360 0–360 0–360

Car 0–40 0–40 0–60
150–215 150–215 100–210
325–360 325–360 320–360
Cup 0–55 0–65 0–90
165–185 150–215 110–215
335–360 325–360 320–360
Vaseline 0–30 0–30 0–40
145–205 140–200 145–300
340–360 315–360 300–360
Table 2. Classification accuracy.
Feature
extraction Generic
method Car Face object
DSIFT 88.7 86.6 88.1

GIST 92.3 90.3 94.2
SURF 90.1 87.3 95.7
The overall performance analysis of the system is shown in Table 2. In this analysis
hand-drawn sketch is used as the query image and Cosine Similarity is used for distance
measurement.
The method was tested using different objects. The query image can be a real image or a
hand-drawn image of the object with a little background as possible. The results show that
the approach is quite stable and is not affected by the choice of the query images. Even the
performance with the hand-drawn sketch is quite good. Out of the three feature extraction
methods GIST gives better detection rate. In the case of rotated object SURF gives good
performance than others.
6 CONCLUSION
Object detection refers to find the position of a particular object in a given image. There are
many object detection methods, mostly based on the training process. The target object for
713

object detection task is defined not only by the sample database but also by human experi-
ence, which is rarely included in the training-based approach. The proposed system describes
a training-free generic object detection method using a single query image. The query image
can be become a typical real image, a virtual image, or even the hand-drawn sketch of the
object of interest. The detection process is similar to template matching. SIFT, GIST and
SURF descriptors are used for feature extraction. In order to determine significant features,
dimensionality reduction is used. The Euclidean distance and MCS are used as the similarity
measurement with a two-step decision approach. The system was tested for car detection,
face detection, generic object detection and rotated object detection. Even if the system uses
only a single query image without training, the results were very good, especially with hand-
drawn sketches as the query images. But for rotated objects, the detection is possible only
upto a certain degree of rotation.
REFERENCES
[1] Bin Xiong and Xiaoqing Ding, “A Generic Object Detection Using a Single Query Image Without
Training”, Tsinghua Science and Technology, April 2012, 17(2): 194–201.
[2] Mikolajczyk K. and Schmid C., “A performance evaluation of local descriptors”, IEEE Trans.
Pattern Analysis and Machine Intelligence, 2005, 27(10): 1615–1630.
[3] Jurie F. and Triggs B., “Creating efficient codebooks for visual recognition”, In: Proceedings of
IEEE International Conference on Computer Vision. Beijing, China, 2005.
[4] Seo H, Milanfar P., “Training-free, generic object detection using locally adaptive regression kernels”,
IEEE Trans. Pattern Analysis and Machine Intelligence, 2010, 32(9).
[5] Shechtman E. and Irani M., “Matching local self-similarities across images and videos”, In:
Proceedings of IEEE Conference on Computer Vision and Pattern Recognition, 2007.
[6] Ferrari V., Tuytelaars T. and Van Gool L., “Object detection by contour segment networks”, In:
Proceedings of European Conference on Computer Vision. Graz, Austria, 2006.
[7] Aude Oliva and Antonio Torralba, “Building the gist of a scene: the role of global image features in
recognition”, Progress in Brain Research, 2006, Vol. 155, ISSN 0079-6123.
[8] Ivan Sikiric, Karla Brkic and Sinisa Segvic, “Classifying traffic scenes using the GIST image
descriptor”, Proceedings of the Croatian Computer Vision Workshop, 2013.
[9] Bay H., Tuytelaars T. and Van Gool L., “Surf: Speeded up robust features”, Computer Vision,
Springer Berlin Heidelberg, pp. 404 417, 2006.
[10] Baofeng Zhang, Yingkui Jiao, Zhijun Ma, Yongchen Li and Junchao Zhu, An Efficient Image
Matching Method Using Speed Up Robust Features, Proceedings of 2014 IEEE International
Conference on Mechatronics and Automation August 3–6, Tianjin, China.
[11] K. Murphy, A. Torralba, D. Eaton and W. Freeman, Object detection and localization using local
and global features, Towards Category-Level Object Recognition, 2005, Vol. 4170: 382400.
[12] Ivan Sikiric, Karla Brkic, Sinisa Segvic, Classifying traffic scenes using the GIST image descriptor,
Proceedings of the Croatian Computer Vision Workshop, 2013.
[13] Agarwal S, Awan A, Roth D, Learning to detect objects in images via a sparse, part-based
representation, IEEE Trans. Pattern Analysis and Machine Intelligence, 2004, 26(2): 1475–1490.
[14] Rowley H, Baluja S, Kanade T., Neural network-based face detection, IEEE Trans. Pattern Analysis
and Machine Intelligence, 1998, 20(1): 22–38.
714

Non-destructive classification of watermelon ripeness using

acoustic cues
Rajeev Rajan & R.S. Reshma

Department of Electronics and Communication Engineering, Rajiv Gandhi Institute of Technology,
Kottayam, India
ABSTRACT: A major issue in the post-harvest phase of the fruit production sector is the
systematic determination of the maturity level of fruits, such as the ripeness of watermel-
ons. Maturity assessment plays an important role while sorting in packing houses during
the export. This paper proposes a support vector machine-based method for the automated
non-destructive classification of watermelon ripeness by acoustic analysis. Acoustic samples
are collected from ripe and unripe watermelons in a studio environment by thumping on the
surface of watermelons. Sound samples are pre-processed to remove silence regions by fixing
an energy threshold. Pre-processed sound signals are segmented into equal-length frames
sized 200 ms, and Teager Energy Operator (TEO)—based features are extracted. The entire
set of audio samples are divided into a training set with 60% of the total audio samples and
the remaining 40% for testing. A support vector machine—based classifier is trained with
features extracted from the training set. Twenty dimensional feature vectors are computed in
the feature extraction phase and fed into the classification phase. The results show that the
proposed TEO-based method was able to discriminate between ripe and unripe watermelons
with overall accuracy of 83.35%.
1 INTRODUCTION
During the past decade, the inspection of quality and maturity of fruits and vegetables in
harvest and postharvest conditions is highly demanding in the fruit production industry.
Systematic determination of maturity assessment plays an important role in sorting. The
need for automated non-contact techniques in sorting and grading is in high demand by the
industry. Several techniques to measure firmness and quality have been listed in Abbott et al.
(1968); Chen et al. (1993). Usually farmers identify the maturity and quality levels of fruits
using certain indices, such as the number of days after full bloom, flesh color, thumping
sound, fruit shape, fruit size and skin color. Traditional methods may have their own limita-
tions. For example, judging watermelon ripeness, using its apparent properties such as size
or skin color, is very difficult due to its thick skin. The most common way by which people
used to determine the watermelon ripeness was by tapping the skin of the melon and then
judging the ripeness using the reflected sound. If the sound is dense, then the watermelon is
under-ripe, while if the sound is hollow, then the watermelon is ripe. Examples of a ripe and
unripe watermelon are shown in the Figure 1. The use of automated inspection of fruits and
vegetables has increased in recent decades to achieve higher quality sorting before packaging.
Acoustic features are widely used in many applications in day-to-day life Ayadi et al.
(1995); Piyush et al. (2016). In the study of Miller and Delwiche Miller and Delwich (1989),
spectral information and machine vision were used for bruise detection on peaches and
apricots. Hyper-spectral imaging for detecting apple bruises was investigated by Xing and
De Baerdemaeker Xing and De Baerdemaeker (2005). In Abbaszadeh et al. (2011), a non-
destructive method for quality test using Laser Doppler Vibrometery (LDV) technology is
presented. By means of a Fast Fourier Transform (FFT) algorithm and by considering the
715

Figure 1. Ripe and Unripe watermelon.
response signal to excitation signal ratio, vibration spectra of fruit are analyzed to classify
ripeness. In this paper, we propose a method based on the Teager Energy Operator (TEO)
and support vector machines. In the proposed work the Teager energy-based features give
results par with other nondestructive techniques Baki et al. (2010); Diezma-Iglesias et al.
(2004). The rest of the paper is organized as follows. Section 2 discusses the theory of the
Teager energy operator and support vector machines in detail. Section 3 discusses the experi-
mental description of the proposed method along with the description of the dataset. Sec-
tion 4 presents results with analysis, followed by conclusion in Section 5.
2 THEORETICAL BACKGROUND
2.1 Teager Energy Operator (TEO)

The TEO approach was first proposed by Teager and further investigated by Kaiser H.M.
Teager and S.M. (1990). The applications of TEO include speech analysis Teager and Teager
(1983), speech emotion recognition Ayadi et al. (1995), image texture analysis and mechanical
fault detection Liu et al. (2013).
The Teager energy operator is defined as:
ψ {x(t )} = x 2 (t ) − x(t )x(t ) (1)
in the continuous case (where x means the first derivative of x, and x means the second
derivative). In the proposed work, a non-speech application of the TEO feature is demon-
strated. The steps to compute a TEO-based feature are shown in Figure 2. Acoustic samples
are collected by thumping on the surface of watermelon, as shown in Figure 3. Hamming
windowed audio frame is transformed to the frequency domain using an FFT algorithm, and
power spectrum S(i) is computed, followed by a TEO transform, resulting in:
ψ [S (i )] = S 2 (i ) − S (i + 1)(i − 1) (2)
A Mel-scale filter bank is used to filter the spectrum obtained from the TEO processing.
Each filter in the filter bank is a triangle bandpass filter, Hm, which tries to imitate the fre-
quency resolution of the human auditory system Hui et al. (2008).
The outputs of the filter bank are obtained by,
Pm = ψ [S (i ) ⋅ H m (i ) m = 1, 2,… M ] (3)
716

Figure 2. Steps to compute Teager energy
operator based feature.
Figure 3. Collecting acoustic samples by
thumping watermelon.
Figure 4. (a) Thumping sound collected from watermelon; (b) TEO features extracted for one
acoustic sample.
where M is the number of filter banks. Then log compression is applied to the filter bank
output. Finally Discrete Cosine Transform (DCT) applied in order to compress the spectral
information into the low-order coefficients. The feature vector obtained is expressed by:
M
 k (i − 0.5) 
TEOCEP ( k ) = ∑ cos  ⋅ π  ⋅ log( Pi ). (4)
i =1  M 
Thumping sound collected from a riped watermelon and its TEO feature are shown in
Figures 4(a) and (b) respectively.
717

Table 1. Parameters for LIBSVM.
Sl. No. Parameter Libsvmparameter Value
1 Optimal Objective value (Q) obj -16.33

2 Bias term (b) ρ 0.83
3 No. of SV (N) nsv 21
4 Regularization parameter (C) v 0.9523
2.2 SVM classifier

Support Vector Machine (SVM) is a widely used classifier that computes an optimal hyper-
plane and maximizes the margin between two classes of data in the kernel-induced feature
space Vapnik (1998). In the proposed experiment, we used a support vector machine-based
classifier using LibSVM package. SVM classifier with a linear kernel is used in the experi-
mental evaluation. The parameters used in the SVM classifier are listed in Table 1.
3 EXPERIMENTAL DESCRIPTION
3.1 Data collection

In the proposed system, 37 acoustic data samples were collected from 30 watermelons of
a variety called Carolina cross, with different levels of ripeness. It has green skin, red flesh
and commonly produces fruit weighing between 29 and 68 kg. It takes about 90 days from
planting to harvest. Watermelons were labeled as ripe and unripe based on experience by the
human expert. The maturity of the watermelons was determined for labeling purposes by
manually thumping the watermelons and listening to the sound produced. Thumping sounds
were recorded using a Audio-Technica AT2035 Large Diaphragm Studio Condenser micro-
phone connected to a MAC with Pro Tools-9 digital audio workstation.
3.2 Proposed method

In the feature extraction phase, acoustic samples are pre-processed to remove silence region
by fixing an energy threshold. Acoustic samples are segmented into equal-length frame size
of 200 ms. The sampling frequency used is 44,100 Hz. 50% of successive frames are over-
lapped to smoothen the frame-to-frame transition. Each frame is multiplied with a hamming
window in order to keep the continuity Baki et al. (2010). Twenty dimensional TEO-based
feature vectors are extracted, as explained in Section 2.1. The various steps involved in the
process are shown in Figure 5. The extracted TEO features are fed into the trained classifier
to discriminate between ripe and unripe watermelons during the testing phase. The entire
dataset was divided into two subclasses, in such a way that 60% of the acoustic samples could
be used for the training phase and the rest for the testing phase.
4 RESULTS AND ANALYSIS
The confusion matrix of the proposed classification system is given in Table 2. The proposed
method results in an overall classification accuracy of 83.35% with a 60: 40 (training: test-
ing) pattern. The experimental results proved that the proposed TEO-based system outper-
forms the other acoustic-based methods cited in the literature. When the system identified
all the riped watermelons correctly, 66.7% is the system accuracy for the unripe cases. The
experimental results show that TEOCEP features are very effective in capturing human percep-
tion sensitivity and energy distribution, which is more important for discriminating ripe and
unripe watermelons.
718

Figure 5. Proposed method for classification of watermelon ripeness based on TEO feature.
Table 2. Confusion matrix of RIPE/UNRIPE classification.
Training set Testing set
RIPE UNRIPE RIPE UNRIPE
RIPE 10 0 8 0
UNRIPE 0 13 2 4
5 CONCLUSION
A potential method for non-destructive classification of watermelon ripeness using TEO

based features and support vector machines is proposed in this paper. Teager energy opera-
tor based features have been extensively used for various speech processing applications such
as speaker recognition and speech emotion recognition. In the proposed work, we effectively
used TEO based features for a non-speech processing application, to discriminate ripe and
unripe watermelons. In continuation of our research, we would like to focus on fusion of
various features to improve classification accuracy of the proposed system.
REFERENCES
Abbaszadeh, R., A. Rajabipour, H. Ahmadi, M. Delshad, & M. Mahjoob (2011). Assessment of

watermelon quality using vibration spectra. Innovative Computing Technology Communications in
Computer and Information Science 241(2), 21–29.
Abbott, J., G.S. Bachman, R.F. Childers, J. Fitzgerald, & F.J. Matasik (1968). Sonic techniques for
measuring texture of fruit and vegetables. Food technology 5(12), 101–112.
719

Ayadi, E., M.S. Kamel, & F.K. Moataz (1995). Survey on speech emotion recognition: Features,
classification schemes, and databases. Pattern Recognition, Elsavier 44(3), 572–587.
Baki, S.R., M.A. Mohd, I.M. Yassin, H.A. Hassan, & A. Zabidi (2010). Non-destructive classification
of watermelon ripeness using mel-frequency cepstrum coefficients and multilayer perceptrons. Neural
Networks (IJCNN), The International Joint Conference on, 1–6.
Chen, P.,M.J. McCarthy, & R. Kauten (1993).NMR for internal quality evaluation of fruits and
vegetables. Austral. Postharv. Conf. C.S.I.R.O. Division of Horticulture, Paper No. 1119, 355–358.
Diezma-Iglesias, B.,M. Ruiz-Altisent, & P. Barreiro (2004). Detection of internal quality in seedless
watermelon by acoustic impulse response. Biosystems Engineering, Elsevier 88(2), 221–230.
Hui, G., C. Shanguang, & S. Guangchuan (2008). Emotion classification of infant voice based on
features derived from teager energy operator. Congress on Image and Signal Processing.
Liu, H., J. Wangand, & C. Lu (2013, March). Rolling bearing fault detection based on the teager energy
operator and elman neural network. Mathematical Problems in Engineering 10. Miller, B. & M.
Delwich (1989). Color vision system for peach grading. Trans. ASAE, 32(4), 1484–1490.
Piyush, P., R. Rajan, M. Leena, & B. Koshy (2016, February). Vehicle detection and classification using
audio visual cues. in proceedings of Third international conference on signal processing and integrated
networks(SPIN).
Teager, H. & S. Teager (1983). Some observations on vocal tract operation froma fluid flow of view
in: Vocal fold physiology: Biomechanics, acoustic, and phonatory control. The Denver Center for
performing arts 55, 358–386.
Teager, H.M. & S. Teager (1990). Evidence for nonlinear production mechanisms in the vocal tract,.
Hardcastle W.J., Marchal A. (eds) Speech Production and Speech Modelling. NATO ASI Series
(Series D: Behavioural and Social Sciences), Springer, Dordrecht 55, 241–261.
Vapnik, V. (1998). Statistical learning theory. New York Wiley.
Xing, J. & J. De Baerdemaeker (2005). Bruise detection on jonagold apples using hyperspectral imaging.
Post harvest Biol. Technol. 37(2), 152–162.
720

Predominant instrument recognition from polyphonic music

using feature fusion
Roshni Ajayakumar & Rajeev Rajan

Department of Electronics and Communication Engineering, Rajiv Gandhi Institute of Technology,
Kottayam, India
ABSTRACT: In this paper, the fusion of a modified group delay feature and a frame slope
feature is effectively utilized to identify the most predominant instrument in polyphonic
music. The experiment is performed on a subset of the Instrument Recognition in Musical
Audio Signals (IRMAS) dataset. The dataset consists of polyphonic music, with predominant
instruments such as acoustic guitar, electric guitar, organ, piano and violin. In the classifica-
tion phase, a Gaussian Mixture Model (GMM)—based classifier makes the decision based
on the log-likelihood score. The results show that a phase-based feature modified group
delay works more effectively than magnitude spectrum based features, such as Mel-frequency
cepstral coefficients and frame slope. The classification accuracy of 57.20% is reported from
the fusion experiment. The proposed system demonstrates the potential of the fusion of
features in recognizing the predominant instrument in polyphonic music.
1 INTRODUCTION
Music Information Retrieval (MIR) is a growing field of research with lots of real-world
applications, and is applied well in categorizing, manipulating and synthesizing music. Music
Information Retrieval (MIR) mainly focuses on the understanding and usefulness of music
data through the research, development and application of computational approaches and
tools. Automatic instrument recognition, one of the MIR tasks, has a wide range of applica-
tions ranging from source separation to melody extraction. In Computational Auditory Scene
Analysis (CASA), musical instrument recognition and sound source recognition play a vital
role. Predominant instrument recognition refers to the problem where the prominent instru-
ment is identified from a mixture of instruments playing together. In the literature, we could
see that the instrument recognition by source separation is widely studied in many music infor-
mation retrieval applications. Since there are numerous approaches for source separation, such
as polyphonic pitch estimation in Klapuri (2001) or the separation of concurrent harmonic
sounds, it can act as a front-end for the successive monophonic recognition task.
2 RELATED WORK
In the literature, numerous attempts in instrument recognition have been reported for mono-
phonic or polyphonic audio files. Features derived from a Root-Mean-Square (RMS) energy
envelope via Principle Component Analysis (PCA) can be seen in Kaminsky and Materka
(1995). In another approach Eronen and Klapuri (2000), cepstral coefficients combined with
temporal features are used to classify 30 orchestral instruments with several articulation styles.
The group delay-based feature has also been used for automatic instrument recognition in
721

an isolated environment Diment et al. (2013). Line spectral frequencies-Gaussian Mixture
Model (GMM) framework is used for instrument recognition in Yu and Yang (2014).
In a polyphonic environment, instrument recognition using a Non-Negative Matrix Fac-
torization (NMF) based source-filter model with mel-filter bank cepstral coefficients and
GMM is attempted in Heit tola et al. (2009). In Kashino and Nakadai (1998), a system
transcribing random chords of clarinet, flute, piano, trumpet and violin was presented. In
this paper, we propose a fusion of features for recognizing predominant instrument from
polyphonic music.
The outline of the rest of the paper is as follows. Section 3 explains the overview of the
system. The feature extraction task is discussed in Section 4. In Section 5, the performance
evaluation is explained. Analysis of results is described in Section 6 and finally conclusions
are drawn in Section 7.
3 OVERVIEW OF THE PROPOSED SYSTEM
The block diagram of the proposed system is shown in Figure 1. The experiment is conducted
in four phases. In the first phase, a baseline system with MFCC features are used, followed
by phases with frame slope, modified group delay feature and early fusion of these features.
In all the phases, GMM—based classifiers are used. Sixty-four mixture GMMs are trained
for all instrument models, using audio files in an isolated environment. For each instrument
model, the likelihood score is computed for the test audio file and the model which reports
maximum log-likelihood is declared as the decision. It can be formulated mathematically as
finding the target – λi for which the following criteria is satisfied.
M −1
arg max ∑ log  p (Om / λi )  (1)
1≤ i ≤ R
m=0
where Om, λi, M, R, represent the feature vectors, GMM model for an instrument, number of
feature vectors, and the total number of instrument models, respectively.
In the proposed experiments, three features, namely Mel-frequency cepstral coefficients,

Modified Group Delay Feature (MODGDF), and frame slope features are frame-wise com-
puted from the audio file. The steps to compute these features are discussed in subsequent
sections.
Figure 1. The proposed system. λ1 λ2 … λn represent Gaussian mixture models for isolated instruments.
722

4.1 Mel-Frequency Cepstral Coefficients (MFCC)
In order to extract the MFCC feature, the Fourier analysis is performed on each speech
frame, producing short-time discrete Fourier transform coefficients. The coefficients are then
grouped together in critical bands and weighted by a triangular band pass filter called Mel-
spaced filter banks. The MFCC can then be derived by taking the log of the band-passed
frequency response and calculating the Discrete Cosine Transform (DCT).
4.2 Modified Group Delay Feature (MODGDF)

Group delay features have already been employed in numerous speech processing applica-
tions Murthy and Yegnanarayana (2011); Rajan and Murthy (2013a,b, 2017); Rajan et al.
(2017). The group delay function, τ(ejω), is defined as the negative derivative of an unwrapped
Fourier transform phase. The group delay function of minimum phase signals can be com-
puted directly from the signal by Oppenheim and Schafer (1990):
τ ( e jω ) =
( ) ( ) ( ) ( )
X R e jω YR e jω + YI e jω X I e jω
(2)
X (e )
2
jω
where the subscripts R and I denote the real and imaginary parts, respectively. X(ejω) and Y(ejω)
are the Fourier transforms of x[n] and nx[n] respectively. The denominator is replaced by its
spectral envelope to mask the spiky nature. The modified group delay function (MODGD)
τm(ejω) is obtained as:
τ m ( e jω ) =
( ) ( ) ( ) ( )
X R e jω YR e jω + YI e jω X I e jω
(3)
S (e )
2γ
jω
where S(ejω) is the cepstrally—smoothed version of X(ejω). The group delay function and
modified group delay function for the speech frame are shown in Figures 2(a) and (b),
respectively. Modified group delay functions are converted to spectra using DCT, as shown
in Hegde (2005). In the proposed experiment, 13-dimensional modified group delay features
(MODGDF) are computed from the test and target audio files.
Figure 2. (a) Group delay functions computed for a speech frame; (b) Modified group delay functions
computed for the frame in (a).
723

4.3 Frame slope feature
The slope feature is determined by computing the slope of a linear fit to Mel filterbank
energies. A Least Square Fit (LSF) is computed using the algorithm listed in Madikeri and
Murthy (2011). The Mel filterbank slope feature emphasizes the system information better
than do conventional MFCCs.
5 PERFORMANCE EVALUATION
5.1 Dataset
Instrument Recognition in Musical Audio Signals (IRMAS) dataset consists of more than 6,000
musical audio excerpts from various styles with annotations of the predominant instrument
present. They are excerpts of three seconds for 11 pitched instruments. In the proposed experi-
ment, initially Gaussian mixture models are built using monophonic instrument wave files with
1,000 files per instrument. We considered five classes, namely acoustic guitar, electric guitar,
organ, piano and violin. In the testing phase, 250 files of five classes (50 each) are tested against
these models. All audio files are stored in 16-bit stereo WAV format, sampled at 44.1 kHz.

In the front-end, MFCC/MODGDF/frame slope feature sets are computed for every 10 ms
with a window size of 30 ms. The MFCC system is used as a baseline system for compara-
tive study. Initially GMM models are built for each class in the training phase. During the
testing phase, feature vectors computed from the audio files are fed in to the model and the
accumulated likelihood score is computed. Based on the likelihood score, the system ranks
the matching from the best to worst. The experiment is conducted for five classes with 50 test
files each. The performance is evaluated using classification accuracy in which, if the system
output matches with ground truth, classification is treated as correct.
6 RESULTS AND ANALYSIS
The studies conducted by Diment et al. (2013) for monophonic instrument recognition,
motivated to focus on system characteristics in the proposed polyphonic experiment. From
the literature, it can be seen that conventional MFCC features are widely used for timbre
analysis. But, as mentioned earlier, frame slope features have already been proven to be more
effective in emphasizing system characteristics. The mapping of the MFCC and frame slope
feature set into a two-dimensional feature space using PCA is shown in Figure 3. It is worth
noting that classes are better separated in frame slope feature space than in the MFCC.
Figure 3. 2-dimensional mapping of (a) MFCC features; (b) Frame slope features.
724

Table 1. Confusion matrix of MODGD experiment.
A. Guitar E. Guitar Organ Piano Violin %
A. Guitar 35 6 5 3 1 70.0
E. Guitar 8 14 15 1 12 28.0
Organ 24 7 7 9 3 14.0
Piano 8 0 1 41 0 82.0
Violin 0 0 0 6 42 84.0
Table 2. Confusion matrix of frame slope experiment.
A. Guitar 32 3 1 10 4 64.0
E. Guitar 2 37 1 6 4 74.0
Organ 4 15 8 13 10 16.0
Piano 5 3 0 38 4 76.0
Violin 28 2 0 0 20 40.0
Table 3. Confusion matrix of MODGD + frame slope experiment.
A. Guitar 21 26 1 0 2 42.0
E. Guitar 1 44 1 0 4 88.0
Organ 1 25 15 5 4 30.0
Piano 1 18 3 28 0 56.0
Violin 3 11 0 1 35 70.0
Table 4. Overall classification accuracy.
Sl.No. Feature Name Accuracy (%)
1 MFCC 43.20
2 Frame Slope 54.00
3 MODGDF 55.60
4 MODGDF + Frame Slope 57.20
The confusion matrices for frame slope, MODGD and fusion experiments are shown in
Tables 1–3. The overall accuracy is reported in Table 4. From the experiments, we observed
that frame slope features and MODGD feature are giving complementary information. So,
finally we combined the feature set and conducted the experiment.
The results show that the baseline system reports an overall accuracy of 43.2%. While
the fusion of features improved the individual classification accuracy for electric guitar and
organ, it deteriorated the performance for other classes. It is worth noting that overall results
improved, when considering the experiment as a whole. The experimental results demonstrate
the potential of the modified group delay feature in recognizing the predominant instrument
in polyphonic music and related applications.
7 CONCLUSION
Predominant instrument recognition in polyphonic music is addressed in this paper. Three

features namely MFCC, frame slope and modified group delay features, are computed in the
725

experiment. A GMM Table based classifier is used in the classification phase. The perform-
ance is evaluated using a subset of the IRMAS dataset. Five classes are considered, namely
acoustic guitar, electric guitar, organ, piano and violin. The results show that modified group
delay features are very much more effective in predominant instrument recognition in poly-
phonic music as compared to conventional MFCC features. Moreover, the fusion of group
delay feature and slope features is also promising in instrument recognition in a polyphonic
context.
REFERENCES
Diment, A., P. Rajan, T. Heittola, & T. Virtanen (2013). Modified group delay feature for musical
instrument recognition. In Proceedings of 10th Int. Symp. Comput. Music Multidiscip. Res., Marseille,
France, 2013, 431–438.
Eronen, A. & A. Klapuri (2000). Musical instrument recognition using cepstral coefficients and
temporal features. In Acoustics, Speech, and Signal Processing, 2000. Icassp’00. Proceedings. 2000
IEEE International Conference on 2, II753–II756.
Hegde (2005). Fourier transform based features for speech recognition. PhD dissertation, Indian Institute
of Technology Madras, Department of Computer Science and Engg., Madras, India.
Heittola, T., A. Klapuri, & T. Virtanen (2009). Musical instrument recognition in polyphonic audio
using source-filter model for sound separation. In Proceedings of Int. Soc. Music Inf. Retrieval Conf.,
327–332.
Kaminsky, I. & Materka (1995). Automatic source identification of monophonic musical instrument
sounds. in proceedings of the IEEE Int. Conf. on Neural Networks, 185–194.
Kashino & T. Nakadai, Kinoshita (1998). Application of bayesian probability network to music scene
analysis. In Proceedings of the International Joint Conference on AI, CASA workshop, 115–137.
Klapuri, A. (2001, May). Multipitch estimation and source separation by the spectral smoothness
principle. Acoustics, Speech and Signal Processing (ICASSP), 2001 IEEE International Conference
on, 5, 3381–3384.
Madikeri, S.R. & H.A. Murthy (2011). Mel filter bank energy based slope feature and its application to
speaker recognition. in proceedings of National Communication Conference(NCC), 155–175.
Murthy, H.A. & B. Yegnanarayana (2011). Group delay functions and its application to speech
processing. Sadhana 36(5), 745–782.
Oppenheim, A. & R. Schafer (1990). Discrete time signal processing. New Jersey: Prentice Hall, Inc.
Rajan, R., M. Misra, & H.A. Murthy (2017). Melody extraction from music using group delay functions.
International Journal of Speech Technology 20(1), 185–204.
Rajan, R. & H.A. Murthy (2013a). Group delay based melody monopitch extraction from music. in
proceedings of the IEEE Int.Conf. on Audio, Speech and Signal Processing, 186–190.
Rajan, R. & H.A. Murthy (2013b). Melodic pitch extraction from music signals using modified group
delay functions. In proceedings of the Communications (NCC), 2013 National Conference on, 1–5.
Rajan, R. & H.A. Murthy (2017). Music genre classification by fusion of modified group delay and
melodic features. In proceedings of the Communications (NCC), 2017 National Conference on, 1–5.
Yu, L. & Y. Yang (2014). Sparse cepstral codes and power scale for instrument identification. In
Proceedings of 2014 IEEE Int. Conf. Acoust., Speech Signal Process., 7460–7464.
726

A technique for countering the integer boundary spurs

in fractional phase-locked loops with VCO
R. Vishnu, S.S. Anulal & Jenitha Ravi

Broadcast and Communication Group, CDAC Trivandrum, Kerala, India
ABSTRACT: Integer boundary spur is a mechanism of fractional spur creation, caused by

interactions between the RF VCO frequency and the reference frequency or phase detector
frequency. An algorithm was developed which identifies the integer boundary frequencies
and configures the synthesizer in such a way that the generated spurs are offsetted far away
from the loop bandwidth, so that they are attenuated by the loop filter. The algorithm runs
as a sub-routine in the micro-controller which programs the synthesizer, to modify the mul-
tiplier value accordingly. This algorithm was developed and tested on a fractional frequency
synthesizer operating from 700–1000 MHz, even though the method is independent of fre-
quency of operation.
1 INTRODUCTION
Frequency synthesizers (PLL and VCO combinations) that are capable of operating only at
integer multiples of the phase frequency detector frequency are known as integer-N PLLs and
those which can synthesize finer steps or fractions of the phase frequency detector frequency
are known as fractional synthesizers. Fractional synthesizers generates two types of spurious
signals, namely fractional spurs and integer boundary spurs (IBS). Modern PLLs use higher
order ∑∆ modulators to reduce fractional spurs. Integer boundary spurs are caused by inter-
actions between the RF VCO frequency and the harmonics of reference or PFD frequency.
When these frequencies are not integer related, spur sidebands may appear as sidebands on the
VCO output spectrum at offset frequencies that is the fundamental and harmonics of the dif-
ference frequency between an integer multiple of the reference/PFD frequency and the VCO
frequency. They are spurs generated inside a PLL, but the system designer can predict these
spurs and hence can be avoided. If the difference frequency can be increased in such a way
that it is made larger than the loop bandwidth, then they will be filtered off by the loop filter.
2 BLOCK DIAGRAM AND CIRCUIT DESCRIPTION
The block diagram of the frequency synthesizer is shown below:

The major blocks of this synthesizer are:
1. Reference Oscillator
2. Reference Multiplier
3. R, N counters and fractional interpolator
4. Phase frequency detector
5. Voltage Controlled Oscillator
6. Output divider
This is a fractional N PLL circuit with an additional multiplier in the reference path. There
is an output divider outside the PLL loop. The delta sigma modulator in the N-counter
reduces the fractional spurs. The other major spurious contribution is by the IBS.
727

Figure 1. Block diagram.
The frequency range of operation for the synthesizer was from 700 MHz to 1000 MHz
in 6.25 kHz steps. The output divider translates the VCO frequency range of 4300 MHz to
5300 MHz, to the required output frequency range. For each frequency, the optimum multi-
plier and R-counter settings are programmed to the synthesizer.
3 DESCRIPTION OF THE TECHNIQUE
3.1 Description
One of the widely used technique for avoiding the IBS is by changing the reference frequency.
This technique will eliminate the IBS caused due to interactions of the VCO with reference as
well as PFD. But it is more costly and space consuming. There must be an additional circuit
for the clock synthesizer. This circuit is intended for handheld radio application, so there is
space as well as cost constraints. The IBS generation due to interaction with reference cannot
be avoided using this technique.
The algorithm works by computing the IBS offset frequency for each output frequency
for various PFD frequencies. It starts with the highest PFD frequency and computes the
IBS offset. A higher PFD frequency will result in a lower N counter and hence better phase
noise performance. If the calculated offset is found to be greater than 2 MHz, the code
exits the loop and program the PLL multiplier and R counter settings for that PFD fre-
quency. The PFD frequency was constrained within 50 MHz to 120 MHz. The lower limit
of 50 MHz is fixed based on a compromise between spurious performance and phase noise.
The charge pump current must be re-programmed for a lower PFD frequency inorder to
keep the loop dynamics constant. Charge pump current changes inversely with the PFD
frequency, so when the PFD frequency increases, the charge pump current must decrease
and vice versa.
3.2 Flowchart
1. Obtain the possible PFD values for multiplier M and R-counter values from 1 to 10.
2. Start with the maximum PFD frequency.
3. Assign value to the variable fPFD. Calculate the value of fVCO based on the present output
divider value
4. For the desired VCO frequency (fVCO), a formula check is done to evaluate whether this
fVCO will generate boundary spur.
Df1 = Absolute [{Roundup (fVCO /fPFD)} × fPFD - fVCO]
Df2 = Absolute [{Rounddown (fVCO /fPFD)} × fPFD - fVCO]
Df1 and Df2 are the spur offset frequencies from the wanted frequency.
5. Check to see if these values are greater than 2 MHz. If this condition is not met, then go
back and check the spur offsets for the next lower PFD frequency.
6. If the spur offset is greater than 2 MHz, then proceed to configure the PLL for that PFD
frequency. The multiplier and R-counter settings must be programmed.
728

Figure 2. Flowchart.
3.3 Validation of the algorithm

Consider an arbitrary frequency fOUT = 768.025 MHz for spur analysis. The analysis can be
done as per the flowchart sequence.
Reference = 19.2 MHz.
There are 6 possible PFD values based on the various combinations of multiplier M and
R-counter values from 1 to 10.
1. The possible values are 51.2, 67.2, 76.8, 86.4, 96, 115.2 MHz
2. Starting with the maximum value of 115.2 MHz
3. fPFD = 115.2 MHz. Then fVCO = fOUT × DIV = 768.025 × 6 = 4608.15 MHz
4. Df1 = Absolute [{Roundup (4608.15/115.2)} × 115.2 - 4608.15] = 115.05 MHz
Df2 = Absolute [{Rounddown (4608.15/115.2)} × 115.2 - 4608.15] = 0.15 MHz
5. Df1 & Df2 > 2 MHz is NOT satisfied. So proceed with the next lower value of PFD frequency.
6. fPFD = 96 MHz. The fVCO = fOUT × DIV = 768.025 × 6 = 4608.15 MHz
7. Df1 = Absolute [{Roundup (4608.15/96)} × 96 - 4608.15] = 95.85 MHz
Df2 = Absolute [{Rounddown (4608.15/96)} × 96 - 4608.15] = 0.15 MHz
729

Figure 3. Algorithm validation.
8. Df1 & Df2 > 2 MHz is NOT satisfied. So proceed with the next lower value of PFD
frequency.
9. fPFD = 86.4 MHz. The fVCO = fOUT × DIV = 768.025 × 6 = 4608.15 MHz
10. Df1 = Absolute [{Roundup (4608.15/86.4)} × 86.4 - 4608.15] = 57.45 MHz
Df2 = Absolute [{Rounddown (4608.15/86.4) } × 86.4 - 4608.15] = 28.95 MHz
11. Df1 & Df2 > 2 MHz is satisfied. So proceed with programming the PLL with this PFD
frequency.
So in this case if we are programming the PLL with PFD of 115.2 MHz or 96 MHz,
there will be IBS at frequency offsets ±300 kHz from the center frequency. But if the PFD
is made 86.4 MHz, the spur offsets will be moved away to >29 MHz. But this frequency has
an additional problem, since this VCO frequency is also an integer multiple of the reference
frequency of 19.2 MHz (240th multiple of 19.2 MHz) (4608.15 is not integer multiple, 4608
is tehe 240th multiple). This issue is later discussed in the evaluation results section.
The functioning of the algorithm over the entire frequency range of operation was validated
in a spreadsheet software. The Figure 3, shows the spurious frequency offsets at different
frequencies, with the spur offsets from the fundamental on the right y-axis, converging
frequency number in the left y-axis and VCO frequency on the x-axis. It is evident that there
are several frequency points where the spur offsets crosses the 2 MHz bound, where the PFD
is changed and spurs offsets are moved off to a higher value.
4 EVALUATION RESULTS
Figure 4 shows the spectrum taken at the output frequency of 792.025 MHz. Using an out-
put divider of 6, the VCO frequency will be 4752.15 MHz. 4752 MHz is an integer multiple
of 86.4 MHz (55th multiple). 86.4 MHz is one of the choice of PFD frequency for a reference
of 19.2 MHz.
Now applying the algorithm,
Df1 = Absolute [{Roundup (4752.15/86.4)} × 86.4 - 4752.15] = 86.25 MHz

So IBS can be observed as sidebands at 150 kHz offsets. This is seen in the YELLOW
trace.
Now if we re-calculate with 115.2 MHz as PFD frequency,
Df1 = Absolute [{Roundup (4752.15/115.2)} × 115.2 - 4752.15] = 86.25 MHz

Hence the offset frequencies are increased to greater than 28 MHz so that it is easily filtered
OFF by the loop filter. RED trace shows the same frequency at 115.2 MHz PFD frequency.
As mentioned earlier the IBS generation due to interaction of the VCO with the refer-
ence cannot be avoided using this technique. To demonstrate this phenomenon, an output
730

Figure 4. Spectrum of 792.025 MHz.
Figure 5. Spectrum of 768.025 MHz.
frequency of 768.025 MHz maybe taken. The VCO frequency is 768.025 × 6 = 4608.15 MHz.

The frequency 4608 MHz is an integer multiple of 19.2 MHz, 96 MHz as well as 115.2 MHz.
But it is not an integer multiple of 86.4 MHz. The algorithm come up with the final PFD
value of 86.4 MHz. But even then the boundary spurs are not fully suppressed due to the
interaction of VCO with the reference frequency. This phenomenon cannot be evaded in this
architecture, since the reference is not tunable.
This can be seen in the Figure 5 shown above. The yellow trace shows the output with the
PFD frequency as 115.2 MHz and the red trace shows the output spectrum with the PFD
frequency as 86.4 MHz. The reference frequency 19.2 MHz also interacts to produce the
150 kHz and their harmonics and divided down spurious.
REFERENCES
[1] “Wideband Synthesizer with Integrated VCO”, ADF4350 Data Sheet Rev A04/2011, pp. 23–28.
[2] “An avoidance technique for mitigating the integer boundary spur problem in a DDS-PLL hybrid
frequency synthesizer,” 2015 International Conference on Communications and Signal Processing
(ICCSP), Melmaruvathur, 2015, pp. 0443–0446. R. Vishnu and Anulal S. S.
[3] “Analyzing, Optimizing, and Eliminating Integer Boundary Spurs in Phase-Locked Loops with
VCOs at up to 13.6 GHz” Analog Dialogue, Aug 2015, Vol. 49 by Robert Brennan.
[4] “Ultra-Low Noise PLLatinum Frequency Synthesizer With Integrated VCO datasheet (Rev. J)”,
LMX2541 Data Sheet Rev J.
731

Adaptive neuro-fuzzy based base station and relay station

deployment for next generation wireless communication
Mariya Vincent
Department of Electronics and Communication Engineering, Rajagiri School of Engineering
and Technology, Kakkanad, India
ABSTRACT: Improper deployment of Base Stations (BSs) and Relay Stations (RSs) at
inappropriate locations can result in a decrease in power efficiency and throughput and an
increase in deployment cost, transmission delay, transmission loss and power consumption.
Hence, in order to exploit the advantages of BSs and RSs completely, an effective selec-
tion and deployment scheme for BS and RS, which attains the target Coverage Ratio (CR)
and throughput, at an affordable deployment budget is required for next generation wireless
communication. The superiority in performance of the proposed method over conventional
clustering methods is illustrated through MATLAB simulations.
1 INTRODUCTION
Wireless communication have become a crucial part in our daily personnel and professional
lives by exchanging information reliably from anywhere, any time. The enormous increase in
the number of mobile subscribers has proportionally increased the data rate demand. The
mobile subscribers also demand anytime-anywhere wireless broadband services with high
quality of experience. One of the solutions to provide adequate signal-to-noise ratio (SNR),
increase data throughput, overall coverage and system capacity is to decrease cell area and
deploying more number of BSs which escalates the deployment cost as well as increase the
inter-cell interference. Hence alternate intelligent solutions should be implemented in next
generation wireless communication networks.
Multi-hop relay (MHR) network is one of the promising solutions recommended by the
LTE-A standards to satisfy the above mentioned service requirements. Yang et al. (2009)
in LTE-A and IEEE 802.16j standards, the concept of MHR network is introduced, where
RSs are deployed along with BSs to improve the coverage and capacity. RSs are more suited
solution in the locations where the backhaul connection is expensive or unavailable. The RSs
also have other advantages like less carbon dioxide (CO2) emission, less power consumption,
easier and faster installation and low maintenance cost. Unlike the BS, which is connected
to the backhaul by a wired connection, RS can be wirelessly connected to the BS. More
over, MHR networks help to improve network throughput and at the same time covers more
number of mobile users over a larger coverage area. Hence, MHR network is considered as
one of the potential candidate to facilitate power efficient wireless communication. The use
of RSs to extend battery was presented by Laneman and Womell (2000). Cho et al. (2009)
proposed an RS deployment scheme to reduce the delay due to handovers by deploying the
nodes at the boundaries of adjacent cell edges. A two stage joint BS and RS placement (JBRP)
scheme is proposed by Lu and Liao (2009) where the authors used k-supplier concepts in the
first stage to deploy BSs and greedy-heuristic concepts for the RS deployment in the second
stage. Even though, the study considered joint BS and RS deployment problem, it ignores the
trade-off between the CR and deployment cost.
The proposed scheme also suffers from unbalanced network load. Chang and Lin (2014),
have proposed an uniform clustering based BSs and RSs placement scheme for MHR network.
733

In the recent years, the study on deployment strategies based on fuzzy logic has attracted
active attention among academic and industry. Fuzzy logic based BS deployment scheme is
proposed by Jau and Lin (2015). It has been shown that the fuzzy based deployment scheme
is low complex and offers better coverage and throughput performance than the conven-
tional uniform clustering based deployment scheme. Vincent et al. (2017) proposed a power-
aware joint fuzzy based BS and RS deployment scheme for green radio communication. The
authors used CR and traffic ratio (TR) as the input parameters for fuzzy inference engine.
To account the greenness of the system, a novel parameter, power consumption effectiveness
ratio (PCER) is given as the third input metric for the fuzzy inference engine. To maximize
PCER, CR, TR, the fuzzy rules are formed and the corresponding selection factor (SF)
is used to choose the BS and RS candidate locations for deployment. The fuzzy inference
engine maps input characteristics to input membership functions and rules to obtained a
single-valued SF. Such a system uses fixed membership functions and a rule structure that is
essentially predetermined by the user’s interpretation of the characteristics of the input vari-
ables. The shape of the membership functions depends on the parameters and scenario of the
network, and changing these parameters change the shape of the membership function. So
we need a system where the shapes of the membership functions and rules arbitrarily changes
with the scenario. Moreover, by applying adaptive neuro fuzzy logic significant decisions can
be taken which resemble those taken by human thinking and reasoning. In the present study,
an effective BSs and RSs deployment based on adaptive-neuro fuzzy was proposed.
The paper is organized as follows: The MHR network model and background are pre-
sented in section 2. In section 3, Adaptive neuro fuzzy based BSs and RSs deployment and
selection phase is described. Section 4 deals with the MATLAB based simulation of the
proposed MHR network scenario and analyses the system performance. Section 5 concludes
the paper.
2 MHR NETWORK MODEL AND BACKGROUND
MHR network model consist of BSs, RSs and MSs. The MSs are located within the geo-
graphical area are uniformly distributed in a large number. According to the geographic fea-
tures of that particular area, there exist some feasible candidate positions where the RSs
can be deployed. The RSs are deployed by the network operators at cell edges and coverage
holes to improve the coverage as well as the capacity of the network. Fig. 1 shows the MHR
network model.
In MHR network, the data can be transmitted directly by the BS to the MSs or be relayed
through the RS, which transmits at relatively lesser power. The multi hop transmission proc-
ess from the BS to the MSs through the RS results in a reduced hop distance between a pair of
Figure 1. MHR network model.
734

actively communicating devices at the same time extends the coverage area of the BS. Owing
to the reduced transmission distance, the power consumption and path loss is reduced, which
ultimately results in an improved data throughput. An RS transmit the signal at relatively
low power than the BS and hence requires lower deployment cost. The distance between any
two communication nodes varies with respect to the deployment scenario. As the distance
increases, path loss increases and the effective signal to noise ratio (SNR) decreases and vice
versa. The system opts for a higher order modulation if the distance between stations is less
and will use lower order modulation if the distance is more. Therefore, the distance between
the participating stations plays a vital role in determining the transmission mode and the
data rate of the link. Hence the MHR networks uses an adaptive modulation coding (AMC)
scheme in which the data rates are adaptively allocated according to the channel conditions
between the communicating stations. Seven modulation coding schemes are considered for
a geographical area which divided into seven non-overlapping regions. The seven distances
from the center to the seven regions correspond to seven SNRs. According to the free space
propagation model, the SNR of each link is given by
 p  c  2
SNR = 10 log10  1    (1)
 pn  4π fr  
where pt, pn, f, c and r represent the transmitted power, thermal noise power, center frequency,
velocity of light and the distance between the transmission stations respectively. The modu-
lation schemes, coding ratios and data rates for various distances and received SNRs are
enumerated in Table 1.
Let DB_R denotes the data rate between BS and RS and DR_M indicates the data rate between
RS and MS. The throughput for indirect transmission between BS and MS is given as,
DB _ R ⋅ DR _ M
DB _ R _ M = (2)
DB _ R + DR _ M
The transmission data rate of MS is decided based on throughput oriented scheme which
is given as,
D = Max( DB _ R _ M , DB _M ) (3)
The average system capacity is given as,

N
∑ D( k )
C= k =1
(4)
N
Table 1. AMC transmission mode in Jau and Lin, (2015).
Received Data
Coding SNR rate Distance
Mode Modulation rate (dB) (Mbps) (km)
1 BPSK 1/2 3.0 1.269 3.2

2 QPSK 1/2 6.0 2.538 2.7
3 QPSK 3/4 8.5 3.816 2.5
4 16-QAM 1/2 11.5 5.085 1.9
5 16-QAM 3/4 15.0 7.623 1.7
6 64-QAM 2/3 19.0 10.161 1.3
7 64-QAM 3/4 21.0 11.439 1.2
735

where D(k) is the transmission data rate of the ith MS and N is the total number of MSs in
the geographical area.
3 ADAPTIV NEURO FUZZY BASED BS AND RS DEPLOYMENT

AND SELECTION PHASE
A two-phase effective BS and RS selection and deployment scheme is explored in the present
study to increase the overall power efficiency and system performance. An adaptive neuro
fuzzy based selection scheme makes an adaptive decision for the deployment of nodes from
the candidate positions.
3.1 BS and RS positions selection phase

In the proposed method, we consider CR and TR as the adaptive neuro fuzzy input sets for
each candidate positions of the BS and RS. Moreover fuzzy based selection strategies are
utilized for the deployment of BS and RS in Vincent et al. (2017).
The CR of ith BS and RS candidate position can be calculated as,
1
CRi = ( NCi ) (5)
N
where NCi is the number of MSs covered under ith RS candidate position and N is the
number of MSs in the geographical area.
The fuzzy sets for CR of ith RS candidate positions takes the linguistic variables like low,
medium, high. The corresponding membership function plot is shown in Fig. 2.
TRi of ith RS candidate position is given by,
Ad ,i − At ,i
TRi = (6)
{
max Ad ,i , At ,i }
where Ad,i and At,i are the average data transmission rate and average traffic demand of MSs
covered by the ith RS candidate position respectively. CR and TR are the two fuzzy inputs
which are considered in the conventional fuzzy based BS deployment schemes [Vincent et al.
(2017)].
Similarly, the fuzzy sets for TR of ith RS candidate positions takes the linguistic variables
like negative, center and positive. The corresponding membership function plot is shown in
Fig. 3. The variations of the crisp values associated with the output with respect to inputs
are shown in Fig. 4. It can be seen that the output is low in magnitude for smaller values of
Figure 2. Membership diagram of CR. Figure 3. Membership diagram of TR.
736

Figure 4. Surface plot of Output versus CR-TR combinations.
the combinations and increases with the increase in their values. The results obtained are in
accordance with the fuzzy rules.
3.2 BS an RS deployment phase

The budget constraint for a total deployment budget (TDB) is expressed as,
u v
∑ DC
i =1
BS ⋅ β i + ∑ DCRS ⋅ γ i ≤ TDB
i =1
(7)
where DCBS and DCRS is the deployment cost of BS and RS respectively. u and v are the
number of candidate positions of BS and RS respectively. βi and γi are defined as follows:
1, if BS is deployed in the i th candidate position

βi =  (8)
0, otherwise
1, if RS is deployed in the i th candidate position
γi =  (9)
0, otherwise
The coverage constraint is expressed as,
C ≥ ECR (10)
where ECR is the expected CR.

The joint BS and RS deployment scheme based on the budget constraint is illustrated as
follows:
Input: Candidate positions of BS, Candidate positions of RS, positions of the MSs, TDB
Output: BS and RS deployment locations, CR of the MSs and average power consumed
Initialization: M = 0 (number of BS positions selected) and temp result = null.
Step 1: Load the training data from workspace to adaptive neuro fuzzy inference engine
Step 2: Set the membership numbers and patterns then test the trained data
Step 3: View the generated membership functions and rules generated.
Step 4: Output is calculated for all the candidate position of BSs.
Step 5: Find out the maximum number of BSs that can be deployed based on the budget
constraint. Identify the candidate locations with highest output for BS deployment
Step 6: Deploy BS in the candidate positions with highest output. Output for all the candi-
date position of RSs in the cell edge of the deployed BSs is calculated.
Step 7: Select the RS candidate location with highest output and check for budget constraint
in (7).
737

Step 8: If the condition in (7) is satisfied, go to step 6. Otherwise, go to step 4 and continue
until either the budget constraint is satisfied or all the RSs within the coverage of
selected BS are deployed.
Step 9: The temp result stores the final positions of BSs and RSs. In addition to that, the
average CR, and average throughput per user of MSs are calculated.
The joint BS and RS deployment scheme based on coverage constraint is illustrated as
follows:
Input: Candidate positions of BSs, Candidate positions of RSs, positions of the MSs,
expected CR (ECR).
Output: BS and RS deployment locations, CR of the MSs and average power consumed.
Initialization: M = 0 (number of BS positions to be selected) and temp result = null.
Step 1: Load the training data from workspace to adaptive neuro fuzzy inference engine
Step 2: Set the membership numbers and patterns then test the trained data
Step 3: View the generated membership functions and rules generated
Step 4: Output is calculated for all the candidate position of BSs.
Step 5: Select the BS with highest output crisp value. Output for all the RSs at the cell edge
of selected BS is calculated.
Step 6: Select the RS with highest output and check for coverage constraint
Step 7: If the coverage constraint in (10) is satisfied, go to step 6. Otherwise, go to step 3 and
continue until (10) gets satisfied or all the RSs within the coverage of selected BS are
deployed.
Step 8: Increment M = M + 1 and go to step 2.
Step 9: The temp result stores the final positions of BSs and RSs. In addition to that the
average CR and average throughput per user of MSs are calculated.
4 SIMULATION RESULTS AND ANALYSIS
The performance of the proposed method is analysed using MATLAB 2015a tool. The pro-
posed fuzzy based BS and RS deployment scheme is compared with uniform clustering based
deployment method. Chang, J. and Y. Lin (2014) has been proved that the coverage and
throughput performance of uniform clustering scheme is far better than JBRP scheme. A
practical wireless cellular network environment is simulated. The assumptions and simula-
tion parameters taken are as follows:
The geographic area is square of size 10 km × 10 km.
The system consists of BSs, RSs and MSs.
The candidate positions of BSs and RSs are randomly selected within the geographic area.
The number of candidate position of BS is taken as 6 and the number of candidate posi-
tion for RSs are varied from 20 to 60.
MSs are uniformly distributed in the geographic area.
The coverage radius of BS and RS are 3.2 km and 1.9 km respectively.
The deployment cost of a BS and a RS is 9 and 3 units respectively.
The traffic demand for each MS is uniformly selected between 5 and 15Mbps.
The data rate between MS and BS or between MS and RS are calculated based on the Table 1
Fig. 5 and Fig. 6 shows the deployment result obtained for a sample simulation environ-
ment based on budget constraint of 50 units and coverage constraint of 80% respectively.
From the simulation results, it can be inferred that the BSs and RSs are judiciously deployed
such that maximum MSs are covered. In Fig 5, the proposed scheme achieves a coverage of
more than 80% at a budget cost of 42 units, which is calculated using [7] and is less than the
target budget. In Fig. 6, more number of BSs and RSs are deployed in order to achieve the
expected coverage of 80% without any budget constraint.
Fig. 7 shows the average throughput comparison between the proposed fuzzy based
scheme and uniform clustering schemes for the deployment budget of 50 units. It is observed
738

Figure 5. Network deployment scenario for Figure 6. Network deployment scenario for a
a budget of 50 units. expected coverage of 80%.
Figure 7. Average throughput per user vs.

Number of candidate position for the RSs Figure 8. Average CR vs. Number of candidate
with TDB of 50 Units. position for the RSs with TDB of 50 Units.
Figure 9. Average throughput per users vs.

Number of candidate positions for the RSs with Figure 10. Average CR vs. Number of candidate
ECR of 80%. position for the RSs with ECR of 80%.
that the increase in the number of RSs will increase the throughput per user initially. But it
is noticed that after certain number of RSs, the average system throughput remains constant
due to the co-channel interference between RSs and between BSs and RSs. Fig. 8 shows the
average CR for the deployment budget of 50 units.
The increase in the number of candidate positions of BS and the deployment of RSs at the
cell edge will increase the CR. But there is no significant improvement in terms of CR, when
the number of candidate locations of RS is above 50.
Fig. 9 and Fig. 10 shows the comparison results of average throughput per user (Mbps)
and CR for a ECR of 80%. Initially average throughput per user and CR will increase with
739

the increase in the number of RSs. But, it maintains constant after certain number of RSs.
The increase in the number of RSs, increases the interference, which also reduces signal to
interference plus noise ratio (SINR). This makes the system to choose lower burst profiles.
5 CONCLUSION
In this paper, an adaptive neuro fuzzy based joint BS and RS deployment scheme is
proposed for next generation wireless communication. The proposed deployment scheme
is formulated for maximizing the network coverage, throughput and power efficiency of
the system. The performance of the proposed scheme is compared with the conventional
uniform clustering based scheme. Our proposed scheme satisfies both budget and coverage
constraints. The proposed method shows improved performance than conventional scheme
for all the considered combinations. The simulation results prove that the proposed scheme
is computationally simple and sustainable for different channel and path loss conditions.
Since the selection and deployment is carried out based on adaptive neuro fuzzy logic,
the proposed method can be considered as a more suited solution for real-time channel
conditions. An adaptive neuro fuzzy based scheme, which considers more number of input
parameters like interference and CO2 emission can be considered for the future study.
REFERENCES
[1] Akyildiz, I.F, D.M. Gutierrez-estevez, R. Balakrishnan and E. Chavarria-reyes (2014). LTE-
Advanced and the evolution to Beyond 4G (B4G) systems, Physical Communication, Vol. 10,
pp. 31–60.
[2] Chang, B.J., Y.H. Liang and S.S. Su (2015). Analyses of Relay Nodes Deployment in 4G
Wireless Mobile Multihop Relay Networks, Wireless Personnel Communication, Vol. 83, No. 2,
pp. 1159–1181.
[3] Chang, J. and Y. Lin (2014). A clustering deployment scheme for base stations and relay stations in
multi-hop relay networks, Computers and Electrical Engineering, Vol. 40, pp. 407–420.
[4] Fettweis., G., and P. Rost, (2011). Green communications in cellular networks with fixed relay
nodes, Cambridge University Press.
[5] Jau-Yang. C and Lin, Y.S. (2015). An Efficient Base Station and Relay Station Placement Scheme
for Multi-hop Relay Networks. Wireless Personal Communications, Vol. 82, No. 3, pp. 1907–1929.
[6] Laneman, J.N. and G.K. Womell (2000). Energy-Efficient Antenna Sharing and Relaying for
Wireless Networks, IEEE Wireless Communication Networking Conference, pp. 7–12.
[7] Lu, H. and W. Liao (2009). Joint Base Station and Relay Station Placement for IEEE 802.16 j
Networks, IEEE Global Telecommunication Conference, pp. 1–5.
[8] Vincent, M., K.V. Babu, M. Arthi and P. Arulmozhivarman (2017). Corrigendum to “Power-aware
fuzzy based joint base station and relay station deployment scheme for green radio communication”
[J. Sustain. Comput.: Inform. Syst. 13 (2017) 1–14].
[9] Yang, Y., H. Hu, J. Xu and G. Mao (2009). Relay Technologies for WiMAX and LTE-Advanced
Mobile Systems, IEEE Communication Magazine, Vol. 47, pp. 100–105.
740

International Conference on Emerging Trends in
Computer Science (ICETICS)

A survey on sentence similarity based on multiple features
Riswana K. Fathima & C. Raseek

Computer Science and Engineering Department, Government Engineering College,
Palakkad, Kerala, India
ABSTRACT: Semantic Textual Similarity (STS) is to decide the level of semantic compara-
bility between pairs of sentences. STS assumes an essential part in Natural Language Process-
ing errands, which has drawn considerable attention in the field of research in recent years. This
survey discusses multiple features including word alignment-based similarity, sentence vector-
based similarity, and sentence constituent similarity to assess the correctness of sentence pairs.
1 INTRODUCTION
Semantic Textual Similarity (STS) is basically a measure used to compute the similarity
between two textual snippets based on the likeliness of their meaning. Measuring semantic
similarity is a difficult assignment since it is simple express a comparable idea in different
ways. Therefore, STS is a deep natural language understanding problem. STS has been gen-
erally utilized as a part of natural language processing tasks such as machine translation
(MT), summarization, generation, question answering (QA), short answer grading, semantic
search, dialog and conversational frameworks and so on.
Previous researchers on semantic text similarity have been centered on records and para-
graphs, while correlation protests in numerous NLP assignments are writings of sentence length,
for example, Video descriptions, News headlines, and beliefs, etc. In this paper, we examine
semantic similarity between two sentences. Given two input textual snippets, we have to conse-
quently choose a score that demonstrates their semantic similarity. In general, the fundamental
undertaking is to process semantic similarity for the given two English sentences in the range
[0, 5], where the score increments with similarity (i.e., 0 indicates no similarity and 5 demon-
strates indistinguishable). Similarity score with the explanation for some English sentences is
shown in Figure 1. The assessment metric utilized is the Pearson correlation coefficient.
STS is also firmly identified with textual entailment (TE) (Dagan et al., 2006), paraphrase
recognition (PARA) (Dolan et al., 2004) and semantic relatedness. All it differs from its tasks.
Textual entailment recognition is the undertaking of choosing, given two text fragments,
regardless of whether the significance of one text is entailed (can be gathered) from another
text. On account of TE, the equivalence is directional. eg: an auto is a vehicle, however a
vehicle isn’t really an auto.
Paraphrase Recognition is a task expects to distinguish in the event that two sentences
have a similar importance of using different words. Two important aspects of paraphrase is
that: same meaning and different words. These two concepts are quite intuitive, but difficult
to formalize. For example, consider the sentences below,
1. Hamilton Construction Company built the new extension.
2. The new extension was built by Hamilton Construction Company.
From the above example, could identify that (1) and the (2) are actually paraphrased.
Since textual entailment and paraphrase detection catches degrees of meaning overlap rather
than making binary classifications of particular relationships. Essentially, semantic related-
ness communicates an evaluated semantic relationship. It is nonspecific about the possibility
743

Figure 1. Similarity scores with explanation.
of the relationship with contradicting material so far being a contender for a high score. For
example, “night” and “day” are abundantly related yet not particularly comparable.
At the beginning of 2012, many efforts are undertaken to have STS over English sentence
pairs. STS shared undertaking has been held yearly since 2012, which provides a platform to
have new algorithms and models. During this period of time, diverse similarity method and
datasets have been explored. One of such similarity signals emerged is a Sultan’s alignment
based method. Also found that deep learning is becoming an upcoming feature set for such
an evaluation mechanism. All things considered, looks into are going ahead to discover the
best performing feature set.
This paper is sorted out as takes after: Section 2 gives the methods used so far for STS.
Section 3 discusses feature sets for STS and provides a comparison between them. Section 4
provides a discussion on datasets used. Section 5 discusses various future research directions
and Section 6 gives a brief concluding comment.
2 METHOD
The methods used so far can be separated into three general classes: alignment approaches,
vector space approaches, and machine learning approach (Cheng et al., 2016). Figure 2,
shows the classification of methodologies based on approaches used.
Alignment approaches align words or phrases in a sentence pair and after that take similar-
ity measure as the quality or scope of alignments. Vector space approaches speak to sentences
as bag-of-words vectors and here the similarity measure will be the vector similarity. Machine
learning approaches consolidate diverse similarity measures and feature utilizes supervised
machine learning models.
This survey aims to consider the evidence from those three set of categories to measure
semantic text similarity between two sentence pairs. Basically, from sentence pairs, we extract
744

Figure 2. Classification of methodologies based on the approach used.
Figure 3. Classification based on features.
the features called alignment-based similarity features, vector-based similarity features, and
sentence constituent similarity features. From the study it found that the extracted feature set
are then joined through a Support Vector Regression (SVR) model to convey the similarity
score between the sentence sets. Figure 3, depicts the classification of feature sets used.
In the following describe how to extract the different types of features.
3.1 Alignment-based similarity feature

Alignment approaches align words or phrases in a sentence pair and after that take similarity
measure as the quality or coverage of alignments. Among this Sultan aligner is the one of
the open source unsupervised word alignment tool (Sultan et al., 2014). The aligner adjusts
related words in two sentences in light of two sets of properties of the words;
1. Whether they are semantically comparative
2. Whether they happen in comparative semantic contexts in the particular sentences
The previous one settles on choice in light of the data gave by the Paraphrase Database
(PPDB) (Ganitkevitch et al., 2013). On account of the last mentioned, contextual similarity
for a word match is computed as the aggregate of the word similarities for each combine of
words with regards to them.
745

In order to have an overview of Sultans aligner let us denote the sentence being aligned as
S(1) and S(2). Whereas here it just portrays content word alignment. It finds outs the word pair
wi( 1) ∈ S ( 1) and w(j 2 ) ∈ S ( 2 ) such that:
1. wi(1) and w (j2) have non-zero semantic comparability, simWij. The figuring of word similar-
ity simWij is as follows;
i. On the off chance that the two words or their lemmas are indistinguishable, at that
point simW = 1.
ii. If the two words are accessible as a couple in the lexical XXXL corpus of the Para-
phrase Database (PPDB), at that point simW.
iii. For any other word pair, simW = 0.
2. Sometimes semantic context of word pair wi(1) and w (j2) have some similarity, denoted as
( )
simCij implies the contextual similarity. wi(1) , w (j2) is figured as the total of the word simi-
( )
larities for each combine of words with regards to wi(1) , w (j2) .
3. Finally, performs one-to-one word alignment.
This is how Sultan aligner aligns related words to STS score. We can summarize that STS
score will be the extent of aligned content words in two input sentences. Sultan aligner found
a STS framework that has effortlessness, high precision and speed and furthermore demon-
strated that it can be sent with no supervision. Difficulty behind it is inability to demonstrate
the semantics of units bigger than words.
The study revealed that extensive feature could be extracted from word alignment to have
semantic textual similarity. One such plan is proposed by Hanig et al. in ExB Themis that
enforced associate degree alignment rule supported named entities, temporal expressions,
measuring expressions and dedicated negation handling. Like Sultan’s aligner this also fol-
lows a strict sequential order;
1. Align named entities nearer to each other. It aligns all name pairs that share no less than one
indistinguishable token. Eg: Michael in one content and both Michael Jackson and Michael
Schumacher in the other, each carries Michael as the identity token and therefore it gets
aligned.
2. Normalized temporal expressions are aligned only if it shares just on the off chance that it
has a similar purpose of time or a similar time interval (e.g. 13:05 and 1.05 pm).
3. Measurement expressions are adjusted iff they convey same outright esteem set (e.g.
$200 k and $200.000).
4. Arbitrary token sequence is then aligned. Which requires different advances and tedious
as well. Make utilization of synonym-lookups and overlooks case information, accentua-
tion characters and symbols. Matches expressions like long term and long-term and US
and United States. Language assets like WordNet and ConceptNet to acquire data of
synonymy (e.g. does and do), antonymy (e.g. doesn’t and does), and hypernymy.
5. Aligns negations (e.g. You are a Christian. versus In this way you are not a Christian).
6. Staying content words are aligned based on cosine similarity on word2vec vectors.
ExB Themis turned out to be the best multilingual system—Spanish and English. Effec-
tively can be adapted to further languages. Likewise extensive feature extraction from word
alignments is an extremely hearty approach. Though it couldn’t advance improved phrase
similarity computation.
3.2 Vector-based similarity features

The framework extracts two vector-based similarity features in light of the two-word vec-
tors subsequently shaped. So for the given two sentence pairs first it generates two vector
representation. At that point register the cosine similarity between those vectors to have their
similarity score.
The primary word vector is learned by the Skip-Gram framework (Baroni et al. 2014).
Skip-Gram model makes use of current words to predict the surrounding window of context
words. Do the better job for infrequent words.
746

The second type of word vectors uses Latent Dirichlet Allocation (LDA) which is a vector
space model (Liu Y et al., 2015). LDA is an example of the topic model. It explains why some
part of data is similar. In LDA first words gathered into documents and each document is a
mix of the humble number of topics. For example, a document wants to be classified as CAT-
related and DOG-related. Then compare the probability of generating various words namely
milk, meow and kitten then classified to CAT-related. Similarly if words like bone, bark and
puppy then it classifies to DOG-related. These are determined using term co-occurrence.
Gibbs algorithm (Griffiths & Steyvers, 2004) is the one of the main algorithm used to have
topic related classifications.
3.3 Sentence constituent similarity features

Cheng et al., proposed to extract the accompanying arrangements of sentence constituent
similarity features:
• Subject similarity: When it convey the subject similarity between two sentences are of the
same then assign 1 for subject similarity else 0.
• Predicate similarity: Assign predicate similarity 1 if the predicate of two sentences are of
same, otherwise 0.
• Object similarity: Assign object similarity 1 if the object of two sentences are of same,
otherwise 0.
• Complement similarity: Assign complement similarity 1 if the complements of two sen-
tences are of same, otherwise 0.
• Named entity similarity: First, it checks whether they are aligned time sets, area sets or
individual matches between looked at sentences. If suppose area could be aligned between
the given sentence pairs means assign 1 for the named entity similarity, otherwise 0.
• Keyword similarity: Assign 1 as keyword similarity view of the proportion of aligned key-
words, otherwise 0.
4 DISCUSSION ON DATASETS
The data set used for the task is of a combination of different settings. The survey discusses
three different settings. Five sets of the dataset in (Cheng et al., 2016) consist of dataset
namely Questions and answers (Q&A) Answer-Answer data set, Headlines data set, Plagia-
rism Detection data set, Post Edited Machine Translations data set and Q&A Question-
Question data set. These datasets were taken from the previous four years (2012–2015) of the
SemEval English STS assignment.
Sultan et al. alignment mechanism in (Sultan et al., 2014) was based on six datasets in par-
ticular deft-forum, deft-news, headlines, images, OnWN and tweet-news. These mentioned
datasets are from SemEval STS 2014. The sentences were gathered from a variety of sources
like discussion forums, news articles, and news headlines and so on. Consist each dataset with
a minimum of 750 sentence pairs.
ExB Themis (Hanig et al., 2015) is a multilingual STS system for English and Spanish lan-
guages. Therefore dataset on English and Spanish are utilized for the task. For English acces-
sible dataset proposed by Agirre et al. in 2012–2014 are utilized. These datasets are taken from
2012–2014. The data set collection in (Agirre et al., 2012) are namely MSRpar, MSRvid, OnWN,
SMTnews, and SMTeuroparl. Similarly, dataset in (Agirre et al., 2013) comprises of HDL,
FNWN, OnWN, SMT and TYPED. Whereas in (Agirre et al., 2014) it is of namely HDL,
OnWN, OnWN, Deft-news, Images and Tweet-news. ExB Themis formed a dataset from above
mentioned as a new combination which comprises of the domain namely forum, students, belief,
headlines and images. For Spanish has taken datasets from Wikipedia and Newswire. Table 1,
shows the overview of the feature sets discussed so far and the datasets used by them in a nutshell.
The performance on each data set is evaluated using the metric called Pearson Correlation
Coefficient. It found that (Cheng et al., 2016) proved a mean of 0.69996 Pearson correlation.
747

Table 1. Overview of the compared features and the datasets used.
Proposed approaches
Types of features Model Data set
Alignment based similarity features F. Cheng et al. Recent years (2012–2015)

SemEval English STS task
Sultan et al. Dataset in SemEval STS 2014
Hanig et al. Available SemEval English
dataset from 2012–2014
Vector-based similarity features F. Cheng et al. Recent years (2012–2015)
Sentence constituent similarity features F. Cheng et al. Recent years (2012–2015)
Sultan aligner (Sultan et al., 2014) found a weighted mean Pearson correlation between
0.7337 to 0.7610. In ExB Themis (Hanig et al., 2015) found a mean of 0.7942 Pearson cor-
relation for English dataset and a mean of 0.6725 Pearson correlation for Spanish data.
5 FUTURE RESEARCH DIRECTIONS
Semantic Textual Similarity (STS) drawn a considerable attention in recent years. It reveals
that STS can be improved further to a greater extent. So far discussed sentence similarity of
monolingual ones. Therefore it shows an open idea to extend the STS in multilingual seman-
tic similarity.
The existing study shows that it is difficult to assign semantics of units larger than words.
In this way, it is an ideal opportunity to create calculations that can adjust to necessities
looked by different information areas and application.
6 CONCLUSION
Semantic Textual Similarity measures the extent of similarity between two sentences. Differ-
ent features including alignment-based similarity features, vector-based similarity features,
and sentence constituent similarity features are used to have the semantic score between the
sentences. Pearson correlation coefficient is used as the evaluation metric. Sentence similarity
based on support vector regression shows the best performance among all other evaluation
strategies with the feature sets.
REFERENCES
Agirre, E., Banea, C., Cardie, C., Cer, D.M., Diab, M.T., Gonzalez-Agirre, A., … & Wiebe, J. (2014).
SemEval-2014 Task 10: Multilingual Semantic Textual Similarity. SemEval@ COLING, 81–91.
Agirre, E., Banea, C., Cardie, C., Cer, D.M., Diab, M.T., Gonzalez-Agirre, A., … & Rigau, G. (2015).
SemEval-2015 Task 2: Semantic Textual Similarity, English, Spanish and Pilot on Interpretability.
SemEval@ NAACL-HLT, 252–263.
Agirre, E., Cer, D., Diab, M., Gonzalez-Agirre, A., & Guo, W. (2013). sem 2013 shared task: Semantic
textual similarity, including a pilot on typed-similarity. In* SEM 2013: The Second Joint Conference
on Lexical and Computational Semantics. Association for Computational Linguistics.
Agirre, E., Diab, M., Cer, D., & Gonzalez-Agirre, A. (2012). Semeval-2012 task 6: A pilot on semantic
textual similarity. In Proceedings of the First Joint Conference on Lexical and Computational Semantics-
Volume 1: Proceedings of the main conference and the shared task, and Volume 2: Proceedings of the Sixth
International Workshop on Semantic Evaluation, Association for Computational Linguistics, 385–393.
748

Bar Haim, R., Dagan, I., Dolan, B., Ferro, L., Giampiccolo, D., Magnini, B., & Szpektor, I. (2006). The
second pascal recognising textual entailment challenge.
Bär, D., Biemann, C., Gurevych, I., & Zesch, T. (2012). Ukp: Computing semantic textual similarity by
combining multiple content similarity measures. In Proceedings of the First Joint Conference on Lexi-
cal and Computational Semantics-Volume 1: Proceedings of the main conference and the shared task,
and Volume 2: Proceedings of the Sixth International Workshop on Semantic Evaluation, 435–440.
Baroni, M., Dinu, G., & Kruszewski, G. (2014). Don’t count, predict! A systematic comparison of
context-counting vs. context-predicting semantic vectors. In ACL (1), 238–247.
Blei, D.M., Ng, A.Y., & Jordan, M.I. (2003). Latent dirichlet allocation. Journal of machine Learning
research, 3(Jan), 993–1022.
Corley, C., & Mihalcea, R. (2005, June). Measuring the semantic similarity of texts. In Proceedings
of the ACL workshop on empirical modeling of semantic equivalence and entailment. Association for
Computational Linguistics, 13–18.
Dagan, I., Dolan, B., Magnini, B., & Roth, D. (2009). Recognizing textual entailment: Rational, evalu-
ation and approaches. Natural Language Engineering, 15(4), i-xvii.
Fu, C., An, B., Han, X., & Sun, L. (2016). ISCAS_NLP at SemEval-2016 Task 1: Sentence Similarity
Based on Support Vector Regression using Multiple Features. SemEval@ NAACL-HLT, 645–649.
Ganitkevitch, J., Van Durme, B., & Callison-Burch, C. (2013, June). PPDB: The Paraphrase Database.
HLT-NAACL, 758–764.
Han, L., Kashyap, A.L., Finin, T., Mayfield, J., & Weese, J. (2013). UMBC_EBIQUITY-CORE: Seman-
tic Textual Similarity Systems. * SEM@ NAACL-HLT, 44–52.
Han, L., Martineau, J., Cheng, D., & Thomas, C. (2015). Samsung: Align-and-Differentiate Approach
to Semantic Textual Similarity. SemEval@ NAACL-HLT, 172–177.
Hänig, C., Remus, R., & De La Puente, X. (2015). ExB Themis: Extensive Feature Extraction from
Word Alignments for Semantic Textual Similarity. SemEval@ NAACL-HLT, 264–268.
Islam, A., & Inkpen, D. (2008). Semantic text similarity using corpus-based word similarity and string
similarity. ACM Transactions on Knowledge Discovery from Data (TKDD), 2(2), 10.
King, M., Gharbieh, W., Park, S., & Cook, P. (2016). UNBNLP at SemEval-2016 Task 1: Semantic Tex-
tual Similarity: A Unified Framework for Semantic Processing and Evaluation. SemEval@ NAACL-
HLT, 732–735.
Li, Y., McLean, D., Bandar, Z.A., O’shea, J.D., & Crockett, K. (2006). Sentence similarity based on
semantic nets and corpus statistics. IEEE transactions on knowledge and data engineering, 18(8),
1138–1150.
Liu, Y., Liu, Z., Chua, T.S., & Sun, M. (2015). Topical Word Embeddings. AAAI, 2418–2424.
Mikolov, T., Chen, K., Corrado, G., & Dean, J. (2013). Efficient estimation of word representations in
vector space. arXiv preprint arXiv:1301.3781.
Pilehvar, M.T., Jurgens, D., & Navigli, R. (2013). Align, Disambiguate and Walk: A Unified Approach
for Measuring Semantic Similarity. IACL (1), 1341–1351.
Šarić, F., Glavaš, G., Karan, M., Šnajder, J., & Bašić, B.D. (2012). Takelab: Systems for measuring
semantic text similarity. In Proceedings of the First Joint Conference on Lexical and Computational
Semantics-Volume 1: Proceedings of the main conference and the shared task, and Volume 2: Proceed-
ings of the Sixth International Workshop on Semantic Evaluation. Association for Computational
Linguistics, 441–448.
Srivastava, S., & Govilkar, S. (2017). A Survey on Paraphrase Detection Techniques for Indian Regional
Languages. International Journal of Computer Applications, 163(9).
Sultan, M.A., Bethard, S., & Sumner, T. (2014). DLS @ CU: Sentence Similarity from Word Alignment.
In SemEval@ COLING, 241–246.
Sultan, M.A., Bethard, S., & Sumner, T. (2015). DLS @ CU: Sentence Similarity from Word Alignment
and Semantic Vector Composition. In SemEval@ NAACL-HLT, 148–153.
749

A review of extractive sentence compression
Shirin A. Fathima & C. Raseek

ABSTRACT: Sentence compression makes a sentence shorter whilst preserving its mean-
ing and grammar; it is applicable in different fields such as text summarization. Among the
various methods to shorten sentences, extractive sentence compression is still far from being
solved. Reasonable machine-generated sentence compressions can often be obtained by con-
sidering subsets of words from the original sentence. This paper presents a survey of the
different approaches to extractive sentence compression.
1 INTRODUCTION
Sentence compression is the process of shortening a text whilst keeping the idea of the origi-
nal sentence. It is in such a way that the grammar and structure of the sentence is greatly
simplified, while the underlying meaning and information remains the same. It makes the
necessary to speak by eliminating unnecessary words or phrases. Sentence compression is
also known as text simplification or summarization.
Sentence compression is an important area of research because it is a backbone for differ-
ent Natural Language Processing (NLP) applications. Normally, human languages contain
complex compound constructions which causes difficulties in automatic modification, clas-
sification or processing of human-readable text. Today, sentence compression techniques are
widely used in many industries, mainly as a part of data mining and machine learning. It has
been widely used for displaying on small screens (Corston-Oliver, 2001), such as in television
captions, automatic title generation (Vandeghinste & Pan, 2004), search engines, topic detec-
tion, summarization (Madnani et al., 2007), machine translation, paraphrasing, and so forth.
There are various techniques for sentence compression, including word or phrase removal,
using shorter paraphrases, and common-sense knowledge. There are, primarily, two types
of sentence compression: extractive and abstractive. In the extractive method, objects are
extracted from the source sentence without modifying the objects themselves. The main idea
here is to find the subset of important words that contain the information of the entire sen-
tence. An example of this is key phrase extraction, where the goal is to select individual words
or phrases that are important (without modifying them) to create a short sentence; it should
preserve the meaning of the original sentence. Abstractive sentence compression considers
semantic representation of the sentence in order to make it simpler.
The task of extractive sentence compression is very complex. It is not simply shortening
a sentence; the properties of the original sentence should be preserved. The performance of
a technique depends upon the compression rate, the grammar and keeping the important
words from the original sentence. Various approaches for extractive sentence compression
include: generative noisy channel models (Knight & Marcu, 2002); tree transduction model
(Knight & Marcu, 2002; Cohn & Lapata, 2007, 2009; Yao et al., 2014); structured discrimina-
tive compression model; Long-Short Term Memory (LSTM) (Clarke & Lapata, 2008); ILP
(Yao & Wan, 2017; De Belder & Moens, 2010; Wang et al., 2013). In addition, the different
techniques make use of machine-learning algorithms such as the Maximum Entropy Model
and Support Vector Machines (SVMs).
751

This survey discusses different methodologies of extractive sentence compression and
makes a competitive study by considering the advantages and disadvantages of each. This
study is helpful because sentence compression has a wide range of applications and many of
them favor extractive compression. This survey will assist in the selection of the right tech-
niques for a particular application.
This paper is organized as follows. Section 2 discusses the machine-learning approach,
the decision-tree model, the Integer Linear Programming (ILP) approach, and the LSTM
approach for extractive sentence compression and provides. Section 3 conducts a compara-
tive critical analysis of these approaches. Section 4 discusses the datasets and evaluation
approaches and makes a comparative study between them. Section 5 gives a brief conclusion.
2 LITERATURE REVIEW
Different approaches are put forward for extracting the important words from a sentence in
order to form a new compressed sentence. Figure 1 shows the approaches discussed here for
extractive sentence compression.
2.1 Tree transduction method

One of the common approaches for extracting the important words from a sentence is the tree
to tree transduction method, which typically makes use of Synchronous Context-Free Grammar
(SCFG). Here, the compression is generated by pruning the dependency tree or constituency
tree (Cohn & Lapata, 2009; Knight & Marcu, 2000; Berg-Kirkpatrick et al., 2011; Filippova &
Altun, 2013). In Cohn and Lapata (2009), the source sentence and target sentence are in the form
of trees. Synchronous grammar will license more than one compression for a source tree where
the SCFG is also in the form of a tree. This rule, in the form of tree shown in Figure 2 (Cohn &
Lapata, 2009), helps to eliminate different branches of the source tree in order to generate the
target tree. Each grammar has a score which helps to calculate the score of the candidates.
The noisy channel model and the statistical model are implemented in the work of Knight
et al. (2011). The statistical model makes use of Probabilistic Context-Free Grammar (PCFG)
on the tree representation in order to extract the relevant phrase. Firstly, the input sentence is
parsed as per Collins (1997). Subsequently, various small output trees are hypothesized and
ranked using PCFG.
Figure 1. Classification of methodologies.
Figure 2. Example of SCFG; dotted lines denote variable correspondences, and denotes node deletion.
752

2.2 Approach based on machine learning
A two-stage approach is taken by Galanis and Androutsopoulos (2010) where each stage
utilizes different machine-learning classification techniques. In the first stage, candidates are
generated using the maximum entropy model. Candidates are ranked based on grammatical-
ity and importance. Choosing the appropriate compressed sentence from the candidates is
the second stage of the process and is done using the SVM regression model.
Galanis and Androutsopoulos (2010) also considered the tree structure of the input source
sentence. The candidates are ranked based on the probability estimations made using differ-
ent properties of the edges. Properties such as the Parts Of Speech (POS) tag denote whether
to remove an edge or not, whether to remove a head or a modifier of the edge, and so on.
SVM is applied to the ranked candidates in order to choose the best one. The model is trained
using the training vector (xi, yi) where xi is a feature vector representing a candidate compres-
sion, and yi is a score indicating how good the compression is. 98 different features are used
for this. Finally, the compression which is most similar to the gold compression is selected.
2.3 Modeling with LSTM recurrent neural network

Thus far, the discussed approaches have to wait until the end of the sentence in order to deter-
mine whether a particular word is important or not. In LSTM, decisions can be made at each
time step. Sakti et al. (2015) consider the basic LSTM Recurrent Neural Network (LSTM-RNN)
and perform a few additional modifications. In basic LSTM, the input layer represents the cur-
rent word and the output layer gives the probability representation that determines whether the
word is important or unimportant. The hidden layer gets activated by the weights associated
with that layer, the current input, and any previous hidden activations related to other words.
In order to handle the sparseness of the input representation and make a better decision,
modifications are performed on the basic structure. For this purpose, a continuous represen-
tation layer is introduced between the hidden layer and input layer, which has a connection
to the output layer. Sakti et al. (2015) also propose a method for pretraining on large unsu-
pervised data to learn sentence representation.
A sequence to sequence model is considered by Filippova et al. (2015). Here the sentences
are considered as sequences of zeroes and ones. LSTM is used for the purpose of remember-
ing long-distance dependencies from the input sentence. Probability estimation makes use of
the model from Sutskever et al. (2014). The input sentence is parsed more than once. In the
first pass, the network learns the sentence representation and in later passes it starts to make
predictions on the importance of the words.
2.4 Modeling with ILP

Approaches based on ILP consider an integer-value representation of sentences. The sentences
are represented as integers. Most methods use a binary representation where each word can have
values of either 0 or 1. The work of Yao and Wan (2017) presents a model of deleting unimpor-
tant words by applying grammatical or language constraints on the binary representation of the
input sentence. The methodology considers the input long sentence as x = {x1,..., xn}. The integer
linear representation of x is δ = {δ1,..., δn} where δi £{0,1}; 0 corresponds to the exclusion of that
particular word and 1 represents the inclusion of that particular word in the sentence.
In Yao and Wan (2017), the procedure starts with an initial random bit vector, and the
bits are flipped according to the constraints in Clarke and Lapata (2008), and Yao and Wan
(2017). The constraints are checked at each node (where the sentence is considered as a
dependency tree). Yao and Wan (2017) introduced a top-down, randomized, constrained,
greedy flipping algorithm, which enabled a better scoring function (McDonald, 2006).
3 CRITICAL ANALYSIS
An analysis of the above discussed approaches for extractive sentence compression is con-
ducted as follows by considering the particular contribution of each approach.
753

Word deletion based on tree transduction, Cohn and Lapata (2009) supports an ample
rewrite operation. That is, it is not only deletion specific; we can rearrange the words to form
a new sentence. The advantage of using STSG is that it can model non-isomorphic tree struc-
tures while having efficient inference algorithms. The method introduced can also be used for
structural matching. The grammar rules introduced here are general and can be applied for
more elaborate tree divergences. Moreover, by adopting a more expressive grammar formal-
ism, we can naturally model syntactically complex compressions without having to specify
additional rules. Also, synchronous grammar will provide a large number of compressions
for a given source tree. The decoding algorithm used here can preserve the grammaticality of
the sentence. The tree model provides greater flexibility.
Moving on to the disadvantages of Cohn and Lapata (2009); because this model uses a
parallel corpus to learn, the accuracy of the grammar generated depends on the accuracy of
the corpus. As tree transduction requires a tree-editing operation, designing the tree-editing
operation is complex. There is no guarantee that the induced rules will have good coverage
on unseen trees. Tree fragments containing previously unseen terminals or non-terminals, or
even an unseen sequence of children for a parent non-terminal, cannot be matched by any
grammar productions. In this case, the transduction algorithm fails as it has no way of cov-
ering the source tree, but it can be recovered by adding a new rule. For unseen productions
there is the problem of under compression. There is also the possibility of spurious ambigu-
ity; the situation in which more than one rule produces the same target tree.
As discussed in this survey, the approach of Galanis and Androutsopoulos (2010) is a
two-stage process which enables the consideration of many compression candidates by con-
sidering different constraints. Different ranking models can be used which makes the system
user friendly and the method can be chosen according to the purpose. This method compares
favorably to a state-of-the-art extractive system. Unlike other recent approaches, our system
uses no handcrafted rules. It also uses a trigram scoring function.
The ranking in Galanis and Androutsopoulos (2010) is performed on the basis of gram-
maticality and importance; it gives less importance to the compression rate. In addition,
because it uses more than one machine-learning process, the system is more complex.
One of the main advantages of the LSTM-based approach of Sakti et al. (2015) is that it does
not require a full parse of the input sentence. As well as considering the pretraining method, it
handles a sparseness of data too. With regard to the pretraining method, the compression rate
is near to the human annotation compression rate. As LSTM gives a more accurate prediction
of the importance of the word, it can be applied in many sentence compression techniques.
However, one of the main disadvantages we could take from Sakti et al. (2015) is that of giving
importance to grammar. Sakti et al. (2015) mainly focus on the compression rate and the impor-
tance of the word. A neural network-related method is quite difficult to implement and from which
the basic LSTM gives a compression rate of close to one, that is, no compression took place.
The methodology of Yao and Wan (2017) is a combination of ILP and randomized greedy
flipping techniques. This methodology proposes an effective and simpler approach for sen-
tence compression. Comparing this model with that which uses only ILP, it does not have a
large number of variables. Even though this Randomized C Greedy Flipping (RCGF) algo-
rithm is iterative, it is effective and fast. A preorder traversal of the dependency tree is consid-
ered, which helps to make decisions during the tree traversal. As the random search procedure
is independently proceeding, so the inference algorithm can be implemented by introducing
a parallel implementation. Using this method, a global optimal solution for sentences can be
obtained effectively and the number of local optimal points are small. This property is par-
ticularly significant for shorter sentences with simpler structures. Some of the local solutions
might be close to the global optimum as well. In addition, this method uses a trigram scoring
function that is more advanced. It can be applied with any kind of objective function.
Moving on to the limitations of the approach (Yao and Wan 2017); it requires multiple
random restarts as a part of the iterations. Due to the complex nature of the problem struc-
ture, with different types of constraints, it is difficult to give a formal mathematical analysis for
worst-case convergence bounds. It is difficult to reach an optimal solution for longer sentences
of more than 15 words. Longer and complex sentences attain only sub-optimal solutions.
754

Table 1. Results comparison.
Written corpus Spoken corpus
Compression Compression
Model Methodology rate Accuracy rate Accuracy
Cohn & Lapata Tree 70.4 58.8 75.5 59.5

(2009) transduction (Grammar) (Grammar)
Galanis & ML-based 71.5 60.2 71.7 59.2
Androutsopoulos (Grammar) (Grammar)
(2010)
Sakti et al. (2015) LSTM 81.9 68.8 x 72.0
(pretraining) (Importance) (Importance)
Yao et al. (2017) ILP 71.7 66.4 72.7 66.2
(Grammar) (Grammar)
4 DATASETS AND RESULTS DISCUSSION
There are different datasets available for analyzing different sentence compression tasks.
Most of the approaches discussed in this survey (Cohn & Lapata, 2009; Sakti et al., 2015;
Galanis & Androutsopoulos, 2010; Yao & Wan, 2017) use the same corpora for evaluation
that was annotated by human annotators.
These four papers consider both written corpora and spoken corpora. For written corpora,
they consider sentences from the British National Corpus (BNC) and sentences from the
American News Text Corpus. The spoken corpora were annotations from broadcast news.
Mostly, compression rate and F1 score measures are considered for automatic evaluation.
The grammatical relationship of generated compression against the gold standard compres-
sion is measured by F1 score. Whereas the LSTM-based method of Sakti et al. (2015) does
not focus on grammar; instead, it considers the importance measure along with compression
rate. Table 1 gives the comparison data.
5 CONCLUSION
Basically, longer sentences are simply a waste of memory and time. Therefore, sentence com-
pression is a necessary task. Extractive sentence compression has a wide range of applications
in different domains. It is one of the key activities of text processing in NLP. Google’s Hand-
Fed AI (artificial intelligence) is one of the best examples. It can also be applied in fields such as
phrasal substitution, especially for figurative expressions (Liu and Hwa, 2016). Hence a survey
on extractive sentence compression may be helpful in order to choose the best technique.
There is a wide range of methodologies for extracting the relevant words from a sentence
in order to form a new compressed sentence. Different authors choose different techniques
for extracting and for making a good sentence. We have discussed a few approaches to extrac-
tive sentence compression and made a comparative study of them.
Extractive sentence compression can be improved further. Most of the methods discussed
here are suitable for smaller sentences but less suitable for larger ones. In future, we can train
with data on a large scale. Thus, we can improve the compression accuracy on large sentences
as well. Extractive compression can be modeled using a sampling-based method. It will be
applicable in different areas such as ECG compression.
REFERENCES
Berg-Kirkpatrick, T., Gillick, D. & Klein, D. (2011). Jointly learning to extract and compress. In
Proceedings of the 49th Annual Meeting of the Association for Computational Linguistics: Human
Language Technologies (pp. 481–490). Stroudsburg, PA: Association for Computational Linguistics.
755

Clarke, J. & Lapata, M. (2008). Global inference for sentence compression: An integer linear program-
ming approach. Journal of Artificial Intelligence Research, 31, 399–429.
Cohn, T. & Lapata, M. (2007). Large margin synchronous generation and its application to sentence com-
pression. In Proceedings of the 2007 Joint Conference on Empirical Methods in Natural Language Process-
ing and Computational Natural Language Learning, 28–30 June 2007, Prague, Czech Republic (pp. 73–82).
Cohn, T.A. & Lapata, M. (2009). Sentence compression as tree transduction. Journal of Artificial Intel-
ligence Research, 34, 637–674.
Collins, M. (1997). Three generative, lexicalised models for statistical parsing. In Proceedings of the
Eighth Conference of the European Chapter of the Association for Computational Linguistics
(pp. 16–23). Gothenburg, Sweden: Association for Computational Linguistics.
Corston-Oliver, S. (2001). Text compaction for display on very small screens. In Proceedings of the
NAACL Workshop on Automatic Summarization (pp. 89–98). Stroudsburg, PA: Association for Com-
putational Linguistics.
De Belder, J. & Moens, M.F. (2010). Integer linear programming for Dutch sentence compression. In A.
Gelbukh (Ed.), Computational linguistics and intelligent text processing. CICLing 2010. Lecture notes
in computer science (Vol. 6008, pp. 711–723).
Filippova, K., Alfonseca, E., Colmenares, C.A., Kaiser, L. & Vinyals, O. (2015). Sentence compression
by deletion with LSTMs. In Proceedings of the 2015 Conference on Empirical Methods in Natural
Language Processing, 17–21 September 2015, Lisbon, Portugal (pp. 360–368). Stroudsburg, PA: Asso-
ciation for Computational Linguistics.
Filippova, K. & Altun, Y. (2013). Overcoming the lack of parallel data in sentence compression. In Proceed-
ings of the 2013 Conference on Empirical Methods in Natural Language Processing, 18–21 October 2013,
Seattle, Washington, USA (pp. 1481–1491). Stroudsburg, PA: Association for Computational Linguistics.
Galanis, D. & Androutsopoulos, I. (2010). An extractive supervised two-stage method for sentence com-
pression. In Proceedings of the 2010 Annual Conference of the North American Chapter of the Asso-
ciation for Computational Linguistics: Human Language Technologies (pp. 885–893). Stroudsburg,
PA: Association for Computational Linguistics.
Knight, K. & Marcu, D. (2000). Statistics-based summarization – step one: Sentence compression. In
Proceedings of the Seventeenth National Conference on Artificial Intelligence and Twelfth Conference
on Innovative Applications of Artificial Intelligence, 30 July – 3 August 2000 (pp. 703–710). Palo Alto,
CA: AAAI Press.
Knight, K. & Marcu, D. (2002). Summarization beyond sentence extraction: A probabilistic approach
to sentence compression. Artificial Intelligence, 139(1), 91–107.
Liu, C. & Hwa, R. (2016). Phrasal substitution of idiomatic expressions. In Proceedings of NAACL-
HLT 2016, 12–17 June 2016, San Diego, California (pp. 363–373). Stroudsburg, PA: Association for
Computational Linguistics.
Madnani, N., Zajic, D., Dorr, B., Ayan, N.F. & Lin, J. (2007). Multiple alternative sentence compres-
sions for automatic text summarization. In Proceedings of the 2007 Document Understanding Confer-
ence (DUC-2007) at NLT/NAACL 2007, April 2007, Rochester, New York.
McDonald, R.T. (2006). Discriminative sentence compression with soft syntactic evidence. In Proceedings
of the 11th Conference of the European Chapter of the Association for Computational Linguistics, 3–7
April 2006, Trento, Italy (pp. 297–304). Gothenburg, Sweden: Association for Computational Linguistics.
Sakti, S., Ilham, F., Neubig, G., Toda, T., Purwarianti, A. & Nakamura, S. (2015). Incremental sentence
compression using LSTM recurrent networks. In Proceedings of 2015 IEEE Workshop on Automatic
Speech Recognition and Understanding (ASRU) (pp. 252–258). New York, NY: IEEE.
Sutskever, I., Vinyals, O. & Le, Q.V. (2014). Sequence to sequence learning with neural networks. In Z.
Ghahraman, M. Welling, C. Cortes, N.D. Lawrence & K.Q. Weinberger (Eds.), Advances in Neural
Information Processing Systems 27 (pp. 3104–3112).
Vandeghinste, V. & Pan, Y. (2004). Sentence compression for automated subtitling: A hybrid approach.
In Proceedings of the ACL Workshop on Text Summarization, Barcelona, July 2004 (pp. 89–95).
Wang, H., Zhang, Y. & Zhou, G. (2013). Sentence compression based on ILP decoding method. In G.
Zhou, J. Li, D. Zhao & Y. Feng (Eds.), Natural language processing and Chinese computing. Com-
munications in computer and information science (Vol. 400, pp. 19–29). Berlin, Germany: Springer.
Yao, J.G. & Wan, X. (2017). Greedy flipping for constrained word deletion. In Proceedings of the Thirty-
First AAAI Conference on Artificial Intelligence (AAAI-17) (pp. 3518–3524). Menlo Park, CA: Asso-
ciation for the Advancement of Artificial Intelligence.
Yao, J.G., Wan, X. & Xiao, J. (2014). Joint decoding of tree transduction models for sentence compres-
sion. In Proceedings of the 2014 Conference on Empirical Methods in Natural Language Processing
(EMNLP), 25–29 October 2014, Doha, Qatar (pp. 1828–1833). Stroudsburg, PA: Association for
756

A survey of figurative language detection in social media
P.D. Manjusha & C. Raseek

ABSTRACT: A figurative language often uses words with a meaning different from its literal
meaning. Detection of figurative language helps in computational linguistics and social media
sentiment analysis. The majority of social media comments make use of figurative language for
better expression of user emotions. Figurative language such as simile, humor, sarcasm, and
irony have widespread uses within social media. The purpose of this paper is to survey various
methods for detection of figurative language based on their distinguishing features.
1 INTRODUCTION
Figurative language expresses the feeling of a writer rather than its literal interpretation. In
the literal interpretation, the writer expresses things as such it is. Figurative language offers
a better understanding and analysis of the text than its literal interpretation. In the current
social media arena, people make use of various figurative language like simile, humor, irony
and sarcasm to express their emotions. Identifying features that better distinguish figurative
language is important for the detection process.
Most of the figurative language has an implicit meaning embedded within it. This provides
difficulty in identifying actual sense meant by the figurative language. Each figurative lan-
guage has specific features exclusively determined for them to better distinguish them, which
can be different for different languages. Recognition of this figurative language, whilst also
combining various features that better distinguish them, acts as ‘trending’ in social media.
There exists a wide range of figurative language and this survey focuses on simile, sarcasm,
irony and humor because most micro-blogging platforms make use of these figures of speech
mostly. A simile is a figure of speech that essentially compares two different things with the
help of words such as: like, as, than. They can have implicit or explicit properties mentioned.
For example: Our room feels like Antarctica. Sarcasm usually makes use of words to express
meaning that is the opposite of its actual meaning. It is mainly used to criticize someone’s
feelings. For example: I love the way my sweetheart cheats on me. Irony includes words that
are the opposite of the actual situation. For example: Butter is as soft as a slab of marble.
Humor is also a figure of speech which is used to produce the effect of laughter and to make
things funny. Humor provides the direct implication of the situation. For example: He faces
more problems than a math book has. Identifying features that better distinguish these figures
of speech act as a trending topic in social media.
Figurative language detection is necessary in different applications such as computational
linguistics, social multimedia, and psychology. Text summarization, machine translation sys-
tems, advertisements, news articles, sentiment analysis and review processing systems make
use of figurative language.
This survey mainly focuses on different methods that make use of various distinguishing
features of different figurative language. It also helps to identify features that better distin-
guish each figurative language.
This paper is organized as follows. Section 2 gives a formal definition of the figurative
language detection with an example; also identified are various distinguishing features for
757

different figurative language term. Section 3 discusses various future research directions for
figurative language detection. Section 4 gives brief concluding comments.
2 FIGURATIVE LANGUAGE DETECTION
A figurative language detection system mainly focuses on identification of various figurative

language terms based on certain distinguishing features.
The main objective of the figurative language detection system is defined as: Given an input
sentence, the system aims to identify the figurative language class in which it belongs. The class
can be simile, sarcasm, irony or humor because we are dealing with social media. An important
step in the figurative language detection is dataset collection. After gathering a collection
of tweets, preprocessing of them is needed in order to filter out the actual structure of each
figurative language. Then the features are extracted for each figure of speech for the purpose
of better classification. Extracted features can then be used to train any classifier. Most of
the existing work makes use of the classifier Support Vector Machine (SVM). So, when a new
instance enters the classifier model, it will produce the figurative language class, in which it
belongs as the predicted output. For example, given the sentence: He looks completely like a
zombie, the system produces an output that it is a simile.
Features that can better distinguish simile, sarcasm, irony and humor includes lexical, syn-
tactic and semantic, pragmatic and finally emotion and sentimental features.
2.1 Lexical features

Lexical features include the presence of N-grams. According to this feature words or phrases that
occur frequently are considered. In Joshi et al. (2015), the proposed method is to detect sarcastic
tweets and it makes use of a unigram feature. Frequently occurred unigrams were counted in the
approach. In Thu and New (2017), the proposed method is to detect irony, sarcasm, simile and
humor at the same time based on supervised classification method. The method uses N-gram
as a feature for classification. It considers unigram, bigram and trigram that occur frequently.
Barbieri and Saggion (2014) proposed a method to automatically detect irony and humor.
The method helps to study aspects of frequency imbalance in tweets based on the most fre-
quent and rare words. Frequent and rare words are determined using the American National
Corpus (ANC) frequency data corpora. The presence of frequent words and rare words in a
sentence can cause unexpectedness. This unexpectedness and incongruity can be used to dis-
tinguish irony and humor. This method also considers the length of the sentence as a feature
in which it checks whether the sentence is long or short.
Khokhlova et al. (2016) also consider N-grams for detection of irony and sarcasm auto-
matically. Certain observations are identified based on these obtained N-grams. The method
identifies that a larger number of constructions with negations are found in ironic texts than
sarcastic texts. Sarcastic texts include a larger number of proper names than ironic texts.
Figure 1. Classification of methodologies for figurative language detection based on the features used.
758

A high frequency of verb ‘love’ is found in sarcasm. Sarcastic texts are more egocentric in
that the writer makes use of the pronoun I mostly.
2.2 Syntactic and semantic features

Each figurative language term has a different syntactic structure. A simile consists of differ-
ent components such as tenor, vehicle, event and comparator, Qadir et al. (2015). The tenor
is the subject of comparison, the vehicle is the object of comparison, the event is an action
or state and the comparator includes words such as: like, as, than. Qadir et al. (2016) applied
rules to recognize the syntactic patterns of similes. The rules used include: NP VERB like
NP; ADJ as [a, an] Noun; ADJ like [a, an] Noun.
Parts of speech are an important feature for detection of figurative language. Khokhlova
et al. (2016) used parts of speech as a distinguishing feature for detection of irony and sar-
casm. The method identified the ratio of parts of speech in both corpora. Ironic texts include
the presence of a large number of nouns. It was also observed that other parts of speech are
common in sarcasm. Sarcasm tends to be more emotional due to the presence of adjectives
and adverbs. The hashtag is another feature used by Khokhlova et al. (2016), in which the
hashtag of ironic texts, giving reference to other authors or objects, are more structured.
Ironic texts are more structured than sarcastic texts.
Barbieri and Saggion (2014) consider synonyms and ambiguity as features for detecting
irony and humor. Irony often conveys two messages at the same time and the choice of the
term rather than its synonym is important. The method used identified synonyms of words
in tweets using WordNet. They then calculated their ANC frequencies and sorted them in
a decreasing rank list. This acts as a distinguishing feature for irony and humor. Another
important feature for irony and humor is ambiguity. The method identifies that ironic text
includes a greater arithmetical average of synsets than humor. Ironic tweets present words
with more meaning, whereas humorous tweets present words with less meaning.
2.3 Pragmatic features

Pragmatic features are mainly used to understand other speakers’ intended meaning.
González-Ibánez et al. (2011) reported a method for identifying sarcasm in Twitter tweets.
The authors investigated the impact of lexical and pragmatic features on the task. Pragmatic
features include the use of positive emoticons such as smileys, and negative emoticons such
as frowning faces. Another feature is ToUser, which indicates whether the tweet contains any
marks that are a reply to another tweet. Sarcastic tweets include a larger number of these
pragmatic features than non-sarcastic tweets. Emoticons are mostly found in large numbers
in sarcastic texts. Barbieri and Saggion (2014) also used emoticons as a feature for detection
of irony and humor. An emoticon feature is the number of :), :D, :(, and ;) in a tweet. Humor-
ous text includes emoticons around four times more than ironic texts. Humorous texts are
more explicit than ironic ones and there is a need to understand humorous texts without the
use of explicit signs such as emoticons. Another feature used by Barbieri and Saggion (2014)
indicates the presence of laughter, such as hahah, lol, rofl, imao.
Capitalization is another feature that can distinguish figurative language. Joshi et al. (2015)
used capital letters as a feature for detection of sarcastic texts. Sarcastic texts include a larger
proportion of capital letters than non-sarcastic texts. Khokhlova et al. (2016) added another
feature, interjection. The presence of various interjections in the tweet is taken into consid-
eration. The presence of punctuation acts as a better pragmatic feature. Punctuation includes
the number of commas, exclamation and quotation marks, full stops, semicolons, ellipses
and hyphens. These are included in the method proposed by Barbieri and Saggion (2014).
2.4 Emotion-based and sentimental features

Emotions and sentiment can be used as better features for figurative language detection.
There are eight basic EmoLex (Emotion Lexicon) emotions: anger, anticipation, disgust,
759

fear, joy, sadness, surprise, trust. Sentiment includes positive, negative and neutral polarities.
Emotion-based features, along with polarity features, can improve the efficiency of a detec-
tion system.
Thu and New (2017) proposed a method to detect several figurative language terms at once
from an emotional point of view. It is used for detection of figurative language such as sar-
casm, irony, simile, and humor, while at the same time using multi-class supervised classifica-
tion. Most of the previous work was based on either detecting figurative language separately
or a combination of the two. The method uses a multi-class supervised classification problem
for better detection of figurative language based on emotions. Ensemble Bagging Classifier
and standard SVM are used for classification, thus performing a comparison. This method
focuses on the impact analysis of different features used by the method. Features used in
the approach include word-based, emotion, sentiment, Bag of Sorted Emotion (BOSE) and
BOSE-TFIDF (Term Frequency–Inverse Document Frequency).
Emotion includes eight basic emotions excerpted from the lexicon EmoLex. Sentiment
features includes positive, negative and neutral polarity for each word in the sentence. BOSE
is created from eight basic emotional features and three sentimental features. First, individu-
ally sort the emotional features and sentiment features according to the value they obtained
from each tweet. As a result, a BOSE is obtained that includes a bag of emotions and senti-
ment together. This could better act as a distinguishing feature. BOSE-TFIDF incorporates
document frequency and inverse document frequency along with BOSE.
Khokhlova et al. (2016) proposed a method for automatic detection of irony and sarcasm.
Their method considers eight emotions along with syntactic features for the detection of
irony and sarcasm. EmoLex is used to assign emotion and polarity to the words of tweets.
A comparison with the National Research Council (NRC) Emotion Lexicon was performed.
Frequency lists of sarcasm and irony corpora were obtained primarily and a comparison
with the EmoLex list performed. This helps to provide accurate emotion to corresponding
words and thus is better in distinguishing sarcasm and irony. Observations inferred that sar-
castic texts are more emotional than ironic texts. The method also assigns polarity to each
word in a tweet and obtained the combined polarity of the tweet. Mostly, the polarity of
the sentence will be the polarity of the majority of words. In such a case, sarcastic texts are
slightly more positive than ironic texts. In all cases it won’t be true because sarcastic tweets
can have a change in polarity in different parts. Polarity changes when one moves from the
first part to next part of the text. A sarcastic sentence often starts with a positive sentiment
and ends with negative sentiment or vice versa.
Joshi et al. (2015) made use of the better future of sentiment category. Their method used
the Vader tool in the NLTK natural language processing toolkit and obtained the sentiment
polarity of subparts of the sarcastic sentence. Thus, it was observed that sarcastic sentences
are more negative than non-sarcastic sentences. Explicit incongruity and implicit incongru-
ity can be used for better detection of sarcasm. Explicit incongruity makes use of senti-
ment words of both polarities and thwarted expectations. Implicit incongruity is expressed
through phrases of implied sentiment rather than using polar words. The main advantage of
this method is that it addresses the problem of assigning polarity to a sarcastic sentence due
to change in its sentiment in parts.
Barbieri and Saggion (2014) proposed a method for automatic detection of irony and
humor. The method used the SentiWordNet sentiment lexicon for obtaining the sentiment of
each word. Here, synsets of words were obtained in the sentence and then assigned a senti-
ment score of positive or negative to the words. There are six features in the sentiment group
such as positive sum, negative sum, positive-negative mean, positive-negative gap, positive
single gap and negative single gap. Positive sum and negative sum are the sum of all positive
and negative scores assigned to words in the sentence respectively. Positive-negative mean is
the average of scores obtained from positive sum and negative sum. Positive-negative gap is
the difference between the positive and negative sum. The positive single gap is the difference
between the most positive word and the mean of all the sentiment scores of words in the sen-
tence. Likewise, for the negative single gap, this considers the most negative word. This also
provides a better distinguishing sentiment feature in case of figurative language detection.
760

Qadir et al. (2015) discuss the effect of polarity in similes. Similes often express positive or
negative sentiment toward an entity. This method makes use of sentiment features, such as
component sentiment, which assign polarity scores to the components of a simile. A simile
consists of different parts, such as subject of comparison, object of comparison, event and a
comparator. Another feature is simile connotation polarity which indicates the overall con-
notation of the simile. The method used by Qadir et al. (2015) makes use of the AFINN sen-
timent lexicon and the MPQA subjectivity lexicon in order to assign polarity to the words.
2.5 Datasets for figurative language detection

Use of figurative language over micro-blogging services is increasing day by day. Identifying
and extracting the relevant dataset is necessary for improving the performance of the system.
Similes in social media do not have a built-in dataset. They are extracted from Twitter by
looking for tweets that have words such as as, than and #simile. Then particular rules are
applied to filter out the similes. Tweets having certain syntactic patterns were selected by
Qadir et al. (2015, 2016).
Irony TwBarbieri 2014 and Irony TwReyes 2013 are two datasets for irony, and Sarcasm
Ptacek 2014 and Sarcasm TwRiloff 2013 are two datasets for sarcasm by Khokhlova et al.
Table 1. Overview of the compared features used for detection of several figurative languages.
Proposed approaches
Figurative
Type of feature Model Highlight of features language
Lexical features Thu & New (2017) Unigram, bigram and Simile, sarcasm,
trigrams are considered irony, humor
Joshi et al. (2015) Unigrams Sarcasm
Barbieri & Presence of frequent and Irony and humor
Saggion (2014) rare words
Khokhlova et al. (2016) N-grams Irony and sarcasm
Syntactic and Barbieri & Synonyms and ambiguity Irony and humor
semantic features Saggion (2014)
Qadir et al. (2016) Syntactic rules Simile
Khokhlova et al. (2016) Parts of speech and hashtags Irony and sarcasm
Pragmatic features Barbieri & Emoticons, laughs and Irony and humor
Saggion (2014) punctuation marks
Joshi et al. (2015) Count of capital letters Sarcasm
Khokhlova et al. (2016) Presence of interjections Irony and sarcasm
González-Ibánez Positive and negative Sarcasm
et al. (2011) emoticons, ToUser
Emotion-based and Thu & New (2017) Eight basic emotions from Simile, sarcasm,
sentimental features EmoLex and three irony, humor
sentimental features from
Vader
Khokhlova et al. (2016) Eight basic emotions from Irony and sarcasm
EmoLex
Qadir et al. (2015) Simile component polarity and Simile
simile connotation polarity
Joshi et al. (2015) Polarity of each word, explicit Sarcasm
and implicit incongruity
Barbieri & Polarity of synsets of words, Irony and humor
Saggion (2014) positive sum, negative
sum, positive-negative gap,
positive-negative mean,
positive single gap and
negative single gap
761

(2016). Irony TwBarbieri 2014 includes 10,000 ironic and 40,000 non-ironic instances. Irony
TwReyes 2013 includes 10,000 ironic and 30,000 non-ironic instances. Sarcasm Ptacek
2014 includes 25,000 sarcastic and 75,000 non-sarcastic instances, and TwRiloff 2013 includes
10,000 sarcastic and 40,000 non-sarcastic instances.
The dataset for the combination of all figurative language was created by Thu and New
(2017). It includes balanced and class imbalanced datasets. Balanced datasets include tweets
having hashtags #simile, #sarcasm, #irony and #humor, collected with no restriction on time
or person. The class imbalanced dataset was created by streaming tweets for a particular day.
Because the datasets and evaluation criteria used in the methodologies for intention detec-
tion discussed above are different, a comparative study based on the results is not possible.
Table 1 shows the different features and contributions of different methods according to dif-
ferent features. It also highlights features that better distinguish different figurative language.
3 FURTHER SUGGESTIONS
Identification of occurrence of one figurative language over another can be identified for
better understanding of the text. For example, identification of sarcastic similes, ironic simi-
les and humorous similes. Automatically inferring implicit properties in the sentence can be
done to improve the efficiency of the system. Figurative language detection can be proc-
essed from different angles that make use of cognitive and psycholinguistic information, ges-
tural information, tone and paralinguistic cues. Another suggestion is to build a system that
works efficiently on large texts of the documents. All these are possible directions for future
research in this area.
4 CONCLUSION
Figurative language provides better analysis and understanding of text than its literal inter-
pretation. There are different types of figurative language terms. But in the case of social
media, simile, sarcasm, irony and humor have an impact. Supervised classification methods
are present for automatic detection of figurative language. Identification of features that
better distinguish figurative language helps in improving the efficiency of the system. Each
figurative language has its own features. Through this survey, we have tried to highlight the
distinguishing features of each figurative language. This survey is conducted with the hope of
shedding some light on the different features of the various figures of speech and how they
can be incorporated for the automatic detection of those languages simultaneously.
REFERENCES
Bamman, D. & Smith, N.A. (2015). Contextualized sarcasm detection on Twitter. In Ninth International
AAAI Conference on Web and Social Media (pp. 574–577). New York, NY: AAAI Press.
Barbieri, F. & Saggion, H. (2014). Automatic detection of irony and humour in Twitter. In S. Colton, D.
Ventura, N. Lavrac & M. Cook (Eds.), Proceedings of the Fifth International Conference on Compu-
tational Creativity, Ljubljana, Slovenia, 10–13 June 2014 (pp. 155–162).
Crossley, S.A., Kyle, K. & McNamara, D.S. (2017). Sentiment Analysis and Social Cognition Engine
(SEANCE): An automatic tool for sentiment, social cognition, and social-order analysis. Behavior
Research Methods, 49(3), 803–821.
Davidov, D., Tsur, O. & Rappoport, A. (2010). Semi-supervised recognition of sarcastic sentences in
Twitter and Amazon. In Proceedings of the Fourteenth Conference on Computational Natural Lan-
guage Learning (pp. 107–116). Stroudsburg, PA: Association for Computational Linguistics.
Fellbaum, C. (1998). WordNet: The encyclopedia of applied linguistics. New York, NY: John Wiley &
Sons.
Fersini, E., Pozzi, F.A. & Messina, E. (2015). Detecting irony and sarcasm in microblogs: The role of
expressive signals and ensemble classifiers. In Proceedings of IEEE International Conference on Data
Science and Advanced Analytics (DSAA) 2015 (pp. 1–8). New York, NY: IEEE.
762

González-Ibánez, R., Muresan, S. & Wacholder, N. (2011). Identifying sarcasm in Twitter: A closer
look. In Proceedings of the 49th Annual Meeting of the Association for Computational Linguistics:
Human Language Technologies: Short Papers (Vol. 2, pp. 581–586). Stroudsburg, PA: Association for
Joshi, A., Sharma, V. & Bhattacharyya, P. (2015). Harnessing context incongruity for sarcasm detection.
In Proceedings of the 53rd Annual Meeting of the Association for Computational Linguistics and the
7th International Joint Conference on Natural Language Processing, Beijing, China, 26–31 July 2015
(Short Papers) (pp. 757–762). Stroudsburg, PA: Association for Computational Linguistics.
Khokhlova, M., Patti, V. & Rosso, P. (2016). Distinguishing between irony and sarcasm in social media
texts: Linguistic observations. In 2016 International FRUCT Conference on Intelligence, Social Media
and Web (ISMW FRUCT), St. Petersburg (pp. 1–6). New York, NY: IEEE.
Liebrecht, C.C., Kunneman, F.A. & van Den Bosch, A.P.J. (2013). The perfect solution for detecting
sarcasm in tweets# not. In Proceedings of the 4th Workshop on Computational Approaches to Subjec-
tivity, Sentiment and Social Media Analysis, Atlanta, Georgia, 14 June 2013 (pp. 29–37). Stroudsburg,
PA: Association for Computational Linguistics.
Mihalcea, R. & Pulman, S. (2007). Characterizing humour: An exploration of features in humorous
texts. In A. Gelbukh (Ed.), Computational Linguistics and Intelligent Text Processing. CICLing 2007.
Lecture Notes in Computer Science (Vol. 4394, pp. 337–347). Berlin, Germany: Springer.
Paul, A.M. (1970). Figurative language. Philosophy & Rhetoric, 3(4), 225–248.
Qadir, A., Riloff, E. & Walker, M.A. (2015). Learning to recognize affective polarity in similes. In Con-
ference Proceedings—EMNLP 2015: Conference on Empirical Methods in Natural Language Process-
ing (pp. 190–200). Stroudsburg, PA: Association for Computational Linguistics.
Qadir, A., Riloff, E. & Walker, M.A. (2016). Automatically inferring implicit properties in similes. In
Proceedings of the 2016 Conference of the North American Chapter of the Association for Computa-
tional Linguistics: Human Language Technologies (pp. 1223–1232). Stroudsburg, PA: Association for
Ramteke, A., Malu, A., Bhattacharyya, P. & Nath, J.S. (2013). Detecting turnarounds in sentiment
analysis: Thwarting. In Proceedings of the 51st Annual Meeting of the Association for Computational
Linguistics, Sofia, Bulgaria, 4–9 August 2013 (pp. 860–865). Stroudsburg, PA: Association for Com-
Reyes, A., Rosso, P. & Buscaldi, D. (2012). From humor recognition to irony detection: The figurative
language of social media. Data & Knowledge Engineering, 74, 1–12.
Reyes, A., Rosso, P. & Veale, T. (2013). A multidimensional approach for detecting irony in Twitter. Lan-
guage Resources and Evaluation, 47(1), 239–268.
Riloff, E., Qadir, A., Surve, P., De Silva, L., Gilbert, N. & Huang, R. (2013). Sarcasm as contrast
between a positive sentiment and negative situation. In Proceedings of the 2013 Conference on Empiri-
cal Methods in Natural Language Processing (pp. 704–714). Stroudsburg, PA: Association for Com-
Taylor, J.M. & Mazlack, L.J. (2005). Toward computational recognition of humorous intent. Proceed-
ings of the Annual Meeting of the Cognitive Science Society, 27, 2166–2171.
Thu, P.P. & New, N. (2017). Impact analysis of emotion in figurative language. In Proceedings of 2017
IEEE/ACIS 16th International Conference on Computer and Information Science (ICIS), Wuhan
(pp. 209–214). New York, NY: IEEE.
Veale, T. & Hao, Y. (2010). Detecting ironic intent in creative comparisons. In Proceedings of the
19th European Conference on Artificial Intelligence, ECAI 2010 (pp. 765–770). Amsterdam,
The Netherlands: IOS Press.
763

A survey of various approaches for interpersonal

relationship extraction
M. Nijila & M.T. Kala

ABSTRACT: Recent efforts to digitize literary works emphasize the importance of a

computational route to analyzing large collections of text. To better understand the literary
narrative, most research focuses on modeling it from the viewpoint of events or characters.
Recently, character-centric approaches have been used to extract relationships between peo-
ple, which enables better understanding of the narrative. Different methodologies consider
distinct types of relationships. The purpose of this paper is to survey the various methodolo-
gies that try to infer interpersonal relationships of characters in narratives. The advantages
and limitations of each method are compared and the various datasets used in each method-
ology are discussed in this paper.
1 INTRODUCTION
Interpersonal relationships have long been studied by analysts in several domains to get a
better understanding of the narratives. These relationships exhibit a variety of phenomena
like family, friendship, hostility, and romantic love. We can see that the narratives are a rich
reflection of these relationships and this provides a better medium for the analysis. Two
approaches used for natural language understanding are an event-centric approach and a
character-centric approach.
An event-centric approach tries to understand the narrative based on the events described
within it. These methods aim to demonstrate the given text using sequences of events, their
participants, and the relationships between them. Such a representation is called a ‘script’.
Another representation of the narrative includes frames, plot units, and schemas.
On the other hand character-centric approaches consider the characters involved in the
narrative. Identifying the relationship between the characters is a better approach to narra-
tive understanding. Recent works are focused on creating a structure called social networks,
sometimes called signed networks, to model relationships between characters. These social
networks are constructed based on co-occurrence of characters in conversations, social
events, and so on.
In general, characterizing the nature of relationships between individuals can assist auto-
matic understanding of the text by explaining the actions of people mentioned in the text
and building expectations of their behavior toward others. Modeling relationships has many
real-world applications, such as predicting possible relationships between people using their
posts or messages in social media, personalizing newsfeeds, predicting virality, and suggesting
friends or topics of interest, for a particular user.
Relationship extraction is commonly achieved through supervised or unsupervised meth-
ods. Some may use hybrid approaches for this. These approaches make use of different
machine-learning algorithms for classification, for example, Naive Bayes and Support Vec-
tor Machines (SVMs). Relationship extraction has the potential to use deep learning models
for a better performance.
765

The purpose of this survey is to discuss different methodologies used for relationship
extraction between characters and make a comparison between them. This paper will give an
overall idea of recent advances in this area.
The layout of this paper is as follows: Section 2 presents a formal definition of interper-
sonal relationship extraction and describes various approaches used for extracting relation-
ship between characters; Section 3 presents avenues for future work in this area; Section 4
gives concluding remarks of this work.
2 RELATIONSHIP EXTRACTION BETWEEN CHARACTERS
Many applications in the field of natural language understanding, information extraction,

and information retrieval require understanding of the relationship between the entities
mentioned in a particular document. Here, entities include a wide grouping which may con-
tain people, locations, organizations, quantities, and so on. This survey focuses only on the
methodologies that extract the relationship between the characters. We can formulate the
problem of relationship extraction between characters as: given a narrative X, the goal of a
relationship extraction system is to identify all the characters mentioned in it, and infer the
relationship between these characters. This will provide a better understanding of the narra-
tive. It will help to process a large collection of text in, for example, digital humanity, political
sciences, and so on. Figure 1 shows the hierarchy of classification of various methodologies.
Figure 2 shows the basic idea of the interpersonal relationship extraction system.
2.1 Supervised approaches

Supervised approaches are the most widely used machine-learning methods. They rely on
a training set where domain-specific examples have been tagged. Supervised approaches
require a suitable tagged corpus.
Culotta et al. (2006) proposed an approach for relationship extraction between characters
using a supervised approach. They proposed a methodology that uses an integrated super-
vised machine-learning method combining both contextual and relational patterns to extract
Figure 1. Classification of methodologies for relationship extraction based on the approach used.
Figure 2. Basic idea of interpersonal relationship extraction system.
766

relationships. Traditional algorithms are able to learn relations such as ‘son’ and ‘sibling’ from
local language clues. But to extract relations like ‘cousin’ we need additional information.
The methodology used by Culotta et al. (2006) creates a Conditional Random Field (CRF)
chain to extract relations while simultaneously uncovering important relational patterns that
improve extraction performance. It is limited to biographical text; for example, encyclopedia
articles. It extracts relations with a model using CRFs. The advantage of this methodology
is that it incorporates relational paths as features to the model. Thus it can learn interesting
relational patterns that may have low precision, without reducing extraction performance.
In this system each learned path has a weighting associated with it. Low-precision patterns
may have a lower weighting than high-precision ones. This methodology will produce better
results only when the database contains accurate relational information and the sentence
contains limited contextual clues. This is a major drawback of this methodology.
Chaturvedi et al. (2016) proposed a method for relation extraction that considers the
evolving nature of the relationship between the characters. Most previous work had made
an assumption that relationships between characters are static. But, in actual fact, the rela-
tionship evolves as we proceed through the narrative. Here, the relationship between people
belongs to two classes, either cooperative or non-cooperative. The approach of Chaturvedi
et al. (2016) made use of a collection of fully labeled and partially labeled sentences from a
narrative to train a semi-supervised segmentation framework. Inputs to this model are a nar-
rative summary and a pair of characters appearing in it. The model segments these sequences
of sentences in the narrative into meaningful and non-overlapping segments, which represent
the flow of relationships in the narrative. The proposed methodology assumes each sentence
in a segment associates with a particular relationship state.
Two types of features are used in this methodology: (1) content features; and (2) transition
features. Content features are used to characterize the textual content of the sentences. It
includes action-based features, adverb-based features, lexical-based features and semantic-
parse-based features. Action-based features model the actions affecting two characters by
identifying all the verbs in the sentence. Adverb-based features are used to model the narra-
tor’s bias toward describing the characters. For that, the model extracts all the adverbs that
modify verbs extracted in action-based features. Lexical-based features are used to analyze
the connotation of all words occurring between a pair of characters. Semantic-parse-based
features incorporate a frame-net style parse of the sentence. Transition features are used to
remember the long history of relationship states.
The advantage of this model is that it considers the flow of information between sentences
to get the evolving relationship. The assumption about the type of relationship is a limitation
of this approach. Rather than considering cooperative and non-cooperative relationships, it
is more valuable to consider diverse sorts of relationships.
Srivastava et al. (2016) proposed an approach for inferring polarity of the relationship
between characters. Their methodology was to formulate the problem as a joint structured
problem for each narrative and present a general model that uses linguistic and semantic
features and features based on the structure of the community. The model jointly infers all
relation labels for a pair of characters in a document. Like the previously discussed approach
by Chaturvedi et al. (2016), the methodology of Srivastava et al. (2016) also considers two
types of relationships (cooperative and adversarial). For training and inference, this method-
ology uses a structured perceptron algorithm. It uses a narrative-specific model that allows
differential weightings for features depending on the narrative type.
The features used in this model are text-based and structural features. Text-based features
are the same as the content features used in the model proposed by Chaturvedi et al. (2016).
These text-based features are used to find relationships between a pair of characters in isola-
tion. The approach proposed by Srivastava et al. (2016) is also motivated by the fact that the
relationship between the characters can also be inferred from the relationship between others
in a narrative. Clique, love triangle, Mexican standoff, and common enemy are the triadic
structural features considered in this approach.
This approach considers the type of narrative and this can be considered as a merit of this
approach. It also shows the importance of understanding the relationship between two characters
767

within the context of their relationship with others in the narrative. The approach makes an
assumption that the relationships between characters are static, which is a major shortcoming.
2.2 Unsupervised approaches

Unsupervised approaches try to find hidden structures from unlabeled data. They do not
need any training phase and can, therefore, be directly applied to any kind of data.
Krishnan and Eisenstein (2014) proposed an unsupervised approach for relationship
extraction. This model produces a signed social network from the content exchanged across
the network edges. It incorporates both network structure and linguistic content and con-
nects these signed social networks with address terms. The usage of the address term indicates
the relationship between two parties. Names, cities, and placeholder names such as ‘dude’,
are considered in this approach. Identification of address terms contain several subtasks:
(1) distinguishing addresses from any mention of other individuals; (2) identifying a lexicon
of titles, which either precede name addresses or can be used in isolation; (3) identifying a
lexicon of placeholder names.
The approach proposed by Krishnan and Eisenstein (2014) made use of a semi-automated
method to construct an address term lexicon, by using bootstrapping from the address term
tagger to build candidate lists. Features used in this methodology are address terms, mutual
friends (using the Adamic–Adar metric) and triads. This approach is only useful to induce
asymmetric signed networks and is a limitation of this method.
Chaturvedi et al. (2017) proposed an approach using unsupervised learning of evolving
relationships between literary characters. The methodology proposed by Chaturvedi et al.
(2016) and Srivastava et al. (2016) assumes that relationships are only of two types: coop-
erative or non-cooperative. The proposed methodology in Chaturvedi et al. (2017) relaxes
this condition and assumes that relationships can be of multiple types. This model makes a
Markovian assumption to capture the ‘flow of information’ or the historical context between
individual sentences.
The model proposed in Chaturvedi et al. (2017) is a Globally Aware General Hidden
Markov Model (GHMM), which incorporates a short historical context via Markovian
assumptions as well as a longer global context using two distinct local and global compo-
nents. This is the main advantage of this model.
2.3 Hybrid approaches

Makazhanov et al. (2014) proposed an approach for extracting family relationships from
novels. This methodology combines word-level techniques with an utterance attribution
approach. This model considers the verbal interactions between characters as candidate rela-
tions, and the speakers themselves as their potential arguments. It also incorporates detection
of vocatives – the explicit forms of address used by the characters in a novel.
The methodology used by Makazhanov et al. (2014) contains four stages for relation
extraction: (1) utterance attribution; (2) vocatives detection; (3) relation extraction; (4) rela-
tion propagation. For each candidate–utterance pair, a feature vector is created. Features
include various quantitative and qualitative characteristics. This methodology is a mixture of
heuristic and supervised approaches.
Once all the utterances are recognized the next step is to detect vocatives. Once a basic fam-
ily relation list is created, we can then proceed to select candidate utterances for the vocative
detection task. This task tried both supervised and unsupervised methods.
The advantage of this methodology is that, at each stage, it uses a combination of different
methods. So it produces better results at each level and that contributes to the final result.
A limitation of this method is inaccurate utterance attribution, a major problem with some
vocative utterances.
Devisree and Raj (2016) proposed a hybrid approach to relationship extraction from sto-
ries. This methodology combines the features of unsupervised and supervised learning meth-
ods and also uses some rules to extract relationships.
768

The methodology proposed by them is designed to work similarly to how the human reader
processes a story to find character relationships. Two phases are included in this method. The
first phase is the sentence-classification stage. Sentences which contain mentions of character
pairs are passed to this phase and then it classifies these sentences into corresponding rela-
tionship classes. If phase one fails to classify these sentences into one of the classes then it is
passed to phase two, which is a text semantic-similarity checking stage.
The advantage of this system is that it uses supervised and rule-based approaches to get
fairly accurate output when compared to existing systems.
2.4 Discussion of datasets for relationship extraction

Articles from the online encyclopedia Wikipedia are used to model the relationship between
characters in various approaches. Culotta et al. (2006) sampled 1,127 paragraphs from 271 arti-
cles and labeled a total of 4,701 relation instances. The structure of the Wikipedia articles some-
what simplifies the extraction task, because important entities are hyperlinked within the text.
SparkNotes is another dataset (‘Plot Overviews’) of 300 English novels. This data-
set contained 50% of fully annotated sequences (402 sentences) and 50% partially anno-
tated sequences (containing 390 sentences, of which 201 were annotated). Of all annotated
sentences, 472 were labeled with a cooperative state. AMT (Amazon Mechanical Turk) is
another dataset used for this task. This dataset was used for evaluation only after training on
SparkNotes data. The work done by Chaturvedi et al. (2016) uses the SparkNotes dataset for
training and the AMT dataset for evaluation.
Srivastava et al. (2016) used the CMU (Carnegie Mellon University) Movie Summary Cor-
pus, a collection of movie plot summaries from Wikipedia, along with aligned metadata, and
then set up an online annotation task using the brat tool, resulting in a dataset of 153 movie
summaries, consisting of 1,044 character relationship annotations.
Some approaches used a dataset of movie dialogs, including roughly 300,000 conversa-
tional turns between 10,000 pairs of characters in 617 movies. The advantage of choosing
this dataset is, it not only provides the script of each movie, but also indicates uses this dataset
in their work. The methodology used by Devisree and Raj (2016) was tested on Jane Austen’s
Pride and Prejudice. This novel contains a fairly high number of characters and a rich set of
family relationships, making this a plausible choice for relationship extraction.
Table 1. Overview of the compared approaches and the datasets used.
Proposed approaches
Type of
approach Model Dataset/Domain Extracted relation Results
Supervised Culotta et al. Articles from wikipedia Mother, cousin, friend, F1 = 0.6363
(2006) education, boss, member P = 0.7343
of and rival R = 0.5614
Chaturvedi et al. SparkNotes, AMT Cooperative and F1 = 0.76
(2016) non-cooperative P = 0.76
R = 0.76
Srivastava et al. Dataset of 300 English Cooperative and F1 = 0.805
(2016) novel summaries non-cooperative P = 0.806
R = 0.804
Unsupervised Krishnan and Dataset of movie scripts Inducing the social F1 = 0.83
Eisenstein (2014) function of address terms
Chaturvedi et al. Dataset of 300 English Familial, Desire, Active, F1 = 0.55
(2017) novel summaries Communicative and
Hostile
Hybrid Devisree and Raj Collection of kids’ Parent-Child, Friendship, P = 0.87
(2016) stories No-Relation R = 0.79
Makazhanov et al. Novels Familial relations A = 0.78
(2014)
769

3 FUTURE RESEARCH
Characterizing the relationships between people can be useful for understanding the text.
Most of the methodologies presented so far were a domain-specific approach. So we need
to extend this approach to different domains. The current approaches made several assump-
tions about the types of relationships that can be relaxed in future work. Future work could
also focus on studying asymmetric relationships. Other directions of study could include the
usefulness of varying text modes (genre, number of characters, time period of novels, etc.)
or mining ‘relationship patterns’ from such texts. Recent advances in deep learning can be
used to get a better result in relation extraction between characters. Error analysis shows
that mismatched co-reference labeling is the most common source of errors in the existing
models. In future, these models could be customized to study the various stages of certain
types of relationships. In addition, including more contextual information can improve the
characterization of relationships.
4 CONCLUSION
There are different methodologies for extracting relationships between characters. Com-
monly used approaches for modeling relationships can be classified into two: supervised
and unsupervised. Some methodologies use a mixture of approaches and are called hybrid
approaches. They take advantage of different methodologies to get better results. Some of
the recent approaches for extracting relationships that belong to these three classes are dis-
cussed in this paper. Research in this area still has room for improvement.
REFERENCES
Agarwal, A., Kotalwar, A., Zheng, J. & Rambow, O. (2013). SINNET: Social interaction network
extractor from text. In The Companion Volume of the Proceedings of IJCNLP 2013: System Demon-
strations, 14–18 October 2013, Nagoya, Japan (pp. 33–36).
Augenstein, I., Das, M., Riedel, S., Vikraman, L. & McCallum, A. (2017). SemEval 2017 Task 10:
Science IE - Extracting keyphrases and relations from scientific publications. arXiv:1704.02853.
Bost, X., Labatut, V., Gueye, S. & Linares, G. (2017). Extraction and analysis of dynamic conversational
networks from TV Series.
Brennan, J.R., Stabler, E.P., Van Wagenen, S.E., Luh, W.-M. & Hale, J.T. (2016). Abstract linguistic
structure correlates with temporal activity during naturalistic comprehension. Brain and Language,
157, 81–94.
Chaturvedi, S., Iyyer, M. & Daumé, H., III. (2017). Unsupervised learning of evolving relationships
between literary characters. In Proceedings of the Thirty-First AAAI Conference on Artificial Intel-
ligence (AAAI-17) (pp. 3159–3165). Palo Alto, CA: AAAI Press.
Chaturvedi, S., Srivastava, S., Daumé, H., III & Dyer, C. (2016). Modeling evolving relationships
between characters in literary novels. In Proceedings of the Thirtieth AAAI Conference on Artificial
Intelligence (AAAI-16), 12–17 February 2016, Phoenix, Arizona (pp. 2704–2710). Palo Alto, CA:
AAAI Press.
Culotta, A., McCallum, A., & Betz, J. (2006). Integrating probabilistic extraction models and data min-
ing to discover relations and patterns in text. In Proceedings of the Human Language Technology Con-
ference of the North American Chapter of the Association of Computational Linguistics (pp. 296–303).
Stroudsburg, PA: Association for Computational Linguistics.
Devisree, V. & Raj, P.R. (2016). A hybrid approach to relationship extraction from stories. Procedia
Technology, 24, 1499–1506.
Frunza, O., Inkpen, D. & Tran, T. (2011). A machine learning approach for identifying disease-treatment
relations in short texts. IEEE Transactions on Knowledge and Data Engineering, 23(6), 801–814.
He, H., Barbosa, D. & Kondrak, G. (2013). Identification of speakers in novels. In Proceedings of the
51st Annual Meeting of the Association for Computational Linguistics (Vol. 1, pp. 1312–1320).
Hoffmann, R., Zhang, C., Ling, X., Zettlemoyer, L. & Weld, D.S. (2011). Knowledge-based weak
supervision for information extraction of overlapping relations. In Human Language Technologies:
770

Proceedings of the 49th Annual Meeting of the Association for Computational Linguistics
(pp. 541–550). Association for Computational Linguistics.
Iyyer, M., Guha, A., Chaturvedi, S., Boyd-Graber, J. & Daumé, H., III. (2016). Feuding families and
former friends: Unsupervised learning for dynamic fictional relationships. In Human Language
Technologies: Proceedings of the 2016 Conference of the North American Chapter of the Associa-
tion for Computational Linguistics (pp. 1534–1544). Stroudsburg, PA: Association for Computational
Linguistics.
Krishnan, V. & Eisenstein, J. (2014). You’re Mr. Lebowski, I’m the dude: Inducing address term formal-
ity in signed social networks. arXiv:1411.4351.
Kwon, H., Kwon, H.T. & Yoon, W.C. (2016). An information-theoretic evaluation of narrative complex-
ity for interactive writing support. Expert Systems with Applications, 53, 219–230.
Ling, Y. (2017). Methods and Techniques for Clinical Text Modeling and Analytics (Doctoral disserta-
tion, Drexel University, Philadelphia, PA).
Makazhanov, A., Barbosa, D. & Kondrak, G. (2014). Extracting family relationship networks from
novels. arXiv:1405.0603.
Pichotta, K. & Mooney, R.J. (2016). Learning statistical scripts with LSTM recurrent neural networks.
In Proceedings of the Thirtieth AAAI Conference on Artificial Intelligence, 12–17 February 2016,
Phoenix, Arizona (pp. 2800–2806). Palo Alto, CA: AAAI Press.
Ru, C., Li, S., Tang, J., Gao, Y. & Wang, T. (2017). Open relation extraction based on core depend-
ency phrase clustering. In 2017 IEEE Second International Conference on Data Science in Cyberspace
(DSC) (pp. 398–404). New York, NY: IEEE.
Shaalan, K. & Oudah, M. (2014). A hybrid approach to Arabic named entity recognition. Journal of
Information Science, 40(1), 67–87.
Srivastava, S., Chaturvedi, S. & Mitchell, T.M. (2016). Inferring interpersonal relations in narrative
summaries. In Proceedings of the Thirtieth AAAI Conference on Artificial Intelligence, 12–17 February
2016, Phoenix, Arizona (pp. 2807–2813). Palo Alto, CA: AAAI Press.
771

A survey of morphosyntactic lexicon generation
M. Rahul & S. Shine

Government Engineering College, Palakkad, Kerala, India
ABSTRACT: Morphosyntactic lexicons are the vocabulary of a language and they have
a vital role in the field of Natural Language Processing (NLP). They provide information
about the morphological and syntactic roles of words in a language. There are various meth-
odologies proposed for the generation of highly accurate and large volumes of morphosyn-
tactic lexicons, which are mainly focused on Machine Learning (ML) approaches. The aim of
this survey is to explore various methodologies for morphosyntactic lexicon generation and
discuss their advantages and disadvantages.
1 INTRODUCTION
A lexicon can be referred to as a large collection of information about the vocabulary of a

language. Information in the sense of the lexical category to which each lexeme belongs. Such
a lexical resource often imposes an internal structure called lexical entries such that it can be
easily accessed by machines for various applications. Lexicons are the primary resource a
Natural Language Processing (NLP) application requires at the critical stages of processing.
More precisely, a lexicon is a component of an NLP system (Guthrie et al., 1996) that bears
both semantic and grammatical information of individual words in a language. Such a lexical
resource is used for NLP applications like machine translation, morphological tagging, part-
of-speech tagging, language modeling and dependency parsing (Allen, 1995).
Machine-readable dictionaries can be considered for dealing with NLP applications which
require large amounts of lexical information. Dictionaries such as Collins’ COBUILD Dic-
tionary (Moon, 2007) and Oxford Advanced Learner’s Dictionary (OALD) (Hornby et al.,
1995), among others, are developed for the above-mentioned applications. But a morpho-
logical and syntactic lexicon, or a morphosyntactic lexicon, can do more in a semantic and
grammatical perspective.
The traditional approach to construct such a lexical resource involves large amounts of
human intervention. Deep lexicographic and morphological analysis is required for the con-
struction of an error-free, high-quality, vast lexicon. Such a job is not very easy because the
vocabulary of a language can be very large. The annotation of words, with corresponding
morphosyntactic properties, demands a great deal of time and requires experienced human
annotators. The difficulties involved in such a job again increase with the complexity of the
language we are dealing with. So, whether a language is an isolating, agglutinating or inflect-
ing one will determine how much effort is necessary to accomplish such a task. Such a lexical
resource is not available for every language; the coverage is limited. Because there is a lot of
manual work in the background, such resources will be very expensive to obtain. In addition,
quality and accuracy are the important factors which highly influence the performance of
the NLP applications that require lexical resource at the critical stages. In order to overcome
these difficulties, we require a language-independent approach to address the problem of
morphosyntactic lexicon generation.
Machine Learning (ML) is a powerful tool for addressing many NLP problems. Here also,
we can make use of various techniques in ML to address the problem of morphosyntactic
lexicon generation. Beginning with the supervised approach, it requires a large volume of
773

annotated data at the training phase. But with the help of freely available lexical resources
like Wiktionary (Zesch et al., 2008) we can construct models that are capable of classifying
the instances in a correct manner. Again, issues such as overfitting and large computation
time will affect the model. Unlabeled resources are available in large scale but we possess few
ways to exploit them. Semi-supervised learning is an efficient approach. It makes use of the
concepts of both supervised and unsupervised learning strategies while, at the same time,
demanding less human effort and giving higher accuracy (Zhu, 2005). Semi-supervised learn-
ing begins with a seed set of annotated instances at the training phase. It uses the various
parameters estimated and labels the rest of the instances (which is purely an unsupervised
approach). Constructing a seed lexicon, and obtaining unlabeled instances of the same, is not
a tedious job for any particular language.
The proposed approaches make use of ML techniques to generate morphosyntactic lexi-
cons in a language-independent manner. Thus they can be applied for any language, no mat-
ter how complex it is.
The aim of this survey is to identify the different methodologies for constructing morpho-
syntactic lexicons and discuss their merits and demerits. This study has high importance in
the current scenario because, as already mentioned, many NLP applications require such a
lexical resource at different stages of processing.
This paper is organized as follows: Section 2 gives a formal definition of the problem and
discusses various classification (machine learning) approaches; Section 3 discusses future
research; Section 4 makes brief concluding comments.
2 MORPHOSYNTACTIC LEXICON
2.1 Definition
A typical morphosyntactic lexicon contains the base forms and inflected forms of words
along with their grammatical and semantic information such as grammatical category and
sub-categorization features. Table 1 shows a subset of a morphosyntactic lexicon for the
English language.
The given lexicon consists of the base form and all the inflected forms for the word ‘cry’.
Each lexical entry is associated with certain attributes such as part of speech, number infor-
mation and gender information. If we can determine the lexical level and the morphological,
syntactic, and semantic relation between these words with the help of unlabeled corpora,
then the generation of large volumes of morphologically and syntactically annotated corpora
are possible. In this survey, the methods proposed are all language-independent so that they
can be adapted independently to any language.
2.2 Semi-supervised methods

Semi-supervised methods best exploit the unlabeled corpora that are comparatively easy to
obtain. Less human effort is required while, at the same time, producing higher accuracy. It
also helps to build good classifiers.
Faruqui et al. (2015) proposed an approach in which they used graph-based semi-super-
vised learning. Graph-based semi-supervised learning has a vital importance in the field of
Table 1. Attributes for base form and its inflected forms.
Word types Attributes
Cry POS: VERB …

Cried POS: VERB, VFORM: FINITE, TENSE: PAST …
Crying POS: VERB, VFORM: GERUND, TENSE: PRESENT …
Cries POS: VERB, NUMBER: SINGULAR, DEGREE: THIRD PERSON …
774

machine learning. The underlying concept is that, using a small set of annotated data points,
or labeled data points, large sets of unannotated or unlabeled data points can be labeled.
Because it is a graph-based approach it begins with a graph which is weighted. The assump-
tion is that the edge connected with a large weighting tends to have the same label. In this
way, the attribute of a labeled node can be propagated through the higher weighted edge to
the unlabeled node.
The method proposed makes use of graph-based semi-supervised learning. While deal-
ing with unlabeled instances, a transductive approach called ‘label propagation’ is used. It
begins with a lexical graph (Polguere, 2009) that contains both labeled and unlabeled nodes.
Nodes refer to the word types in a language. The lexical graph here covers the vocabulary of
English. Nodes are connected with edges which are labeled with some features shared among
them. The features are prefix–suffix information (the English language only contains these
two inflections), morphological transformation, and the cluster number to which both words
belong. Using the seed set that contains only labeled nodes, the system can automatically
generate lexicons which are both syntactically and morphologically annotated.
The model estimation is purely a supervised approach in which only labeled nodes are used
for the empirical estimation of various parameters such as attribute vector and weight vector.
Using the parameters learned, the unlabeled nodes are labeled in the label propagation phase
iteratively. The paradigm projection phase finally determines the most appropriate attribute
labels among multiple possible attributes (multi-label classification problem).
The advantage of this method is that, because it is a graph-based approach, the meth-
odology is said to be language-independent. No language-specific assumptions are made
at any stage of processing. Semi-supervised learning requires few words to be annotated. It
produces a highly accurate, large volume of lexicons. The disadvantage is that the construc-
tion of a lexical graph is a somewhat difficult task because a lexical graph almost covers the
target language.
Ahlberg et al. (2014) address the problem of paradigm induction with the help of inflection
tables. An inflection table contains all the inflected forms of words in a language (Halle &
Marantz, 1993). With inflection tables as input, the system produces a generalization of
inflection paradigms and this helps in assigning attributes correctly to the unlabeled lexical
entries based on the generalization information. Also, with the help of an unannotated cor-
pora, the inflection tables can be expanded and produce more generalizations. The notation
for paradigm consists of ordered string subsequences interspersed with variables. The vari-
ables can be instantiated over other words.
The paradigm generalization algorithm finds the common subsequences among various
words using the Multiple Longest Common Subsequence (MLCS) problem (Pietrzak, 2003;
Wang et al., 2011). The MLCS problem is actually an NP-hard problem but this can be solved
heuristically. At the prediction phase the candidate word is matched against patterns. In order
to enhance the matching, additional information is given; for example, part of speech infor-
mation. When multiple matching occurs then a confidence scoring method helps to select the
most appropriate paradigm.
The advantage of this methodology is that it produces human-readable generalizations.
In addition, because it uses abstract paradigm representation, this only considers how the
segments vary within the inflection table entries. The approach is language-independent. The
disadvantage is that it uses suffix matching to build the tables from base form. It may well
work with inflections close to the end of words, as in English, but not for inflections such as
infix and circumfix.
Banea et al. (2008) proposed a bootstrapping method to construct subjectivity lexicons
(Mihalcea et al., 2007). Because it is a semi-supervised approach, subjective words are used as
the seed set. Banea et al. used an online dictionary and a raw corpus to generate a lexicon. At
each step of the bootstrapping iteration, the seed set was expanded with related words occur-
ring in the online dictionary, in response to a well-formulated query. The noise is filtered out
in the filtering step by calculating the similarity between the extracted candidates and the
seed set. For a similarity calculation, Pointwise Mutual Information and Latent Semantic
Analysis (LSA) (Bouma, 2009) are used. The advantage is that this method can be applied
775

to languages with scarce resources. The disadvantage is that this methodology requires an
online dictionary and only focuses on subjectivity lexicons.
2.3 Supervised methods

Durrett and DeNero (2013) proposed a supervised approach to predict the inflected forms
of a lexical item. The method consists of acquiring orthographic transformation rules of
morphological paradigms automatically from labeled examples. For training they used Wik-
tionary data. Wiktionary is a crowdsourced lexical resource that contains the root word and
all its inflected forms. The first step in the learning procedure consists of identifying inter-
pretable transformation rules, which are generated by aligning each base form to all of its
inflected forms. The learning procedure involves three main steps. These are alignment, span
merging, and rule extraction. In the alignment step each inflected form is aligned with the
base form using an iterative edit-distance algorithm (Marzal & Vidal, 1993). The iterative
edit-distance algorithm determines the lowest cost transformation of the base form into its
inflected form using single-character operations such as insertion, deletion and substitution.
In the prediction phase, transformation rules are applied for an unseen base form, which
will generate more candidate inflectional tables. Semi-Markov Conditional Random Fields
(CRF) (Sarawagi & Cohen, 2005) are then trained to apply these rules correctly to unseen
base forms (prediction). The advantage of this method is that it is a data-driven approach
that best exploits the Wiktionary lexical resource.
Ahlberg et al. (2015) proposed paradigm classification in morphology by using a super-
vised learning method. This method also makes use of inflectional tables and Longest
Common Subsequences (LCS) at intermediate steps to address the problem of paradigm
classification. The method makes use of paradigm generalization (Ahlberg et al., 2014) for
generating generalization paradigms, and a Support Vector Machine (SVM) (Hearst et al.,
1998) for classification.
2.4 Unsupervised methods

Unsupervised methods require a large volume of unannotated data sets which can easily be
obtained for almost every language. Soricut and Och (2015) proposed an unsupervised meth-
odology for morphology induction using concept word embeddings. Here, the algorithm
induces morphological transformations between words by learning the rules of those trans-
formations. This method only considers two kinds of morphological transformations, one
through adding a prefix and the other through adding a suffix. This also provides a mecha-
nism to apply these rules to known words, unknown words and rare words. Prefix−suffix
rules for all possible pairs of words are extracted from the vocabulary and train an embed-
ding space for all words. The skip-gram (Cheng et al., 2006; Guthrie et al., 2006) model is
used for training. The advantage of this methodology is that training requires only a mono-
lingual corpus. No language-specific information is required.
2.5 Discussion on datasets

ML approaches require a large volume of resources for training and evaluation. These may
be annotated or unannotated.
Faruqui et al. (2015) used one year of news articles scraped from different sources for the
construction of a lexical graph and clustered these words with the help of an exchange algo-
rithm because the cluster number is one of the edge features.
Durrett and DeNero (2013) and Ahlberg et al. (2015) used inflectional tables scraped from
the Wiktionary lexical resource, which contains lexical information such as meaning, defini-
tion and description.
The online Romanian dictionary was used by Banea et al. (2008) for expanding the subjec-
tivity lexicon by fetching related words in response to well-formulated queries.
776

Table 2. Overview of the compared approaches and the datasets used.
Type of approach Model Dataset
Supervised Durrett & DeNero (2013) Wiktionary

Ahlberg et al. (2015) Wiktionary
Unsupervised Soricut & Och (2015) Wikipedia
WMT-2013 Shared Task
Arabic Gigaword
Semi-supervised Faruqui et al. (2015) News articles scraped from different sources
Banea et al. (2008) Online Romanian dictionary
Ahlberg et al. (2014) Wiktionary
Soricut and Och (2015) used the Wikipedia resource for the English language. For German,
French and Spanish, they used WMT-2013 Shared Task, and the Arabic Gigaword corpus
was used for the Arabic language. Various other manually created datasets are used for train-
ing and evaluation for individual languages.
3 FUTURE RESEARCH
Semi-supervised learning approaches best fit the problem of morphologically and syntacti-
cally annotated lexicon generation if corpora are available for individual languages. Most
of the approaches discussed so far considered only prefix and suffix transformations. Other
morphological transformations should also be considered while constructing the models
because the model should be capable of handling any complex language. If the methodology
is language-independent, then it can be applied to NLP applications in any language.
4 CONCLUSION
ML approaches can obtain higher accuracy and exhibit better performance. The availability
and correctness of the lexical resource during training and evaluation largely influences the
NLP applications.
This survey has mainly focused on various approaches to addressing the problem of lexi-
con generation with syntactic and morphological information using ML techniques such as
supervised, semi-supervised and unsupervised methods in which most of the methodologies
are language-independent. This makes the adaptation of the various methodologies into any
language possible, which, in turn, accelerates the development of various natural language
applications in individual languages. Semi-supervised learning methodologies are efficient in
this area and can generate more accurate and large volumes of morphosyntactic lexicons.
REFERENCES
Ahlberg, M., Forsberg, M. & Hulden, M. (2014). Semi-supervised learning of morphological para-
digms and lexicons. In Proceedings of the 14th Conference of the European Chapter of the Association
for Computational Linguistics (pp. 569–578). Gothenburg, Sweden: Association for Computational
Linguistics.
Ahlberg, M., Forsberg, M. & Hulden, M. (2015). Paradigm classification in supervised learning of
morphology. In Human Language Technologies: The 2015 Annual Conference of the North American
Chapter of the Association for Computational Linguistics (pp. 1024–1029). Stroudsburg, PA: Associa-
tion for Computational Linguistics.
Allen, J. (1995). Natural language understanding. London, UK: Pearson.
777

Banea, C., Mihalcea, R. & Wiebe, J. (2008). A bootstrapping method for building subjectivity lexicons
for languages with scarce resources. In Proceedings of the 6th International Conference on Language
Resources and Evaluation, LREC-08, 26 May – 1 June 2008, Marrakech, Morocco (pp. 2764–2767).
Paris, France: European Language Resources Association.
Bouma, G. (2009). Normalized (pointwise) mutual information in collocation extraction. In Proceed-
ings of the Biennial GSCL Conference 2009 (pp. 31–40).
Cheng, W., Greaves, C. & Warren, M. (2006). From n-gram to skipgram to concgram. International
Journal of Corpus Linguistics, 11(4), 411–433.
Durrett, G. & DeNero, J. (2013). Supervised learning of complete morphological paradigms. In Human
Language Technologies: Proceedings of the 2013 Annual Conference of the North American Chapter
of the Association for Computational Linguistics (pp. 1185–1195). Stroudsburg, PA: Association for
Faruqui, M., McDonald, R. & Soricut, R. (2015). Morphosyntactic lexicon generation using graph-
based semi-supervised learning. arXiv:1512.05030.
Guthrie, D., Allison, B., Liu, W., Guthrie, L. & Wilks, Y. (2006). A closer look at skip-gram modelling.
In Proceedings of the 5th International Conference on Language Resources and Evaluation (LREC-06)
(pp. 1–4). Paris, France: European Language Resources Association.
Guthrie, L., Pustejovsky, J., Wilks, Y. & Slator, B.M. (1996). The role of lexicons in natural language
processing. Communications of the ACM, 39(1), 63–72.
Halle, M. & Marantz, A. (1993). Distributed morphology and the pieces of inflection. In K. Hale & S.J.
Keyser (Eds.), The view from building 20 (pp. 111–176). Cambridge, MA: The MIT Press.
Hearst, M.A., Dumais, S.T., Osuna, E., Platt, J. & Scholkopf, B. (1998). Support vector machines. IEEE
Intelligent Systems and their Applications, 13(4), 18–28.
Hornby, A.S., Wehmeier, S. & Ashby, M. (1995). Oxford advanced learner’s dictionary (5th ed.). Oxford,
UK: Oxford University Press.
Marzal, A. & Vidal, E. (1993). Computation of normalized edit distance and applications. IEEE Trans-
actions on Pattern Analysis and Machine Intelligence, 15(9), 926–932.
Mihalcea, R., Banea, C. & Wiebe, J. (2007). Learning multilingual subjective language via cross-lingual
projections. Annual Meeting Association for Computational Linguistics, 45(1), 976–983.
Moon, R. (2007). Sinclair, lexicography, and the Cobuild Project: The application of theory. Interna-
tional Journal of Corpus Linguistics, 12(2), 159–181.
Pietrzak, K. (2003). On the parameterized complexity of the fixed alphabet shortest common superse-
quence and longest common subsequence problems, Journal of Computer and System Sciences, 67(4),
757–771.
Polguere, A. (2009). Lexical systems: Graph models of natural language lexicons. Language Resources
and Evaluation, 43(1), 41–55.
Sarawagi, S. & Cohen, W.W. (2005). Semi-Markov conditional random fields for information extraction.
In Proceedings of the 17th International Conference on Neural Information Processing Systems (pp.
1185–1192). Cambridge, MA: MIT Press.
Soricut, R. & Och, F.J. (2015). Unsupervised morphology induction using word embeddings. In Human
Language Technologies: Proceedings of the 2015 Annual Conference of the North American Chapter
of the Association for Computational Linguistics (pp. 1627–1637). Stroudsburg, PA: Association for
Wang, Q., Korkin, D. & Shang, Y. (2011). A fast multiple longest common subsequence (MLCS) algo-
rithm. IEEE Transactions on Knowledge and Data Engineering, 23(3), 321–334.
Zesch, T., Müller, C. & Gurevych, I. (2008). Extracting lexical semantic knowledge from Wikipedia and
Wiktionary. In Proceedings of the 6th International Conference on Language Resources and Evalua-
tion, LREC-08, 26 May – 1 June 2008, Marrakech, Morocco (pp. 1646–1652). Paris, France: Euro-
pean Language Resources Association.
Zhu, X. (2005). Semi-supervised learning literature survey. Technical Report 1530, Computer Science,
University of Wisconsin-Madison. Retrieved from https://minds.wisconsin.edu/bitstream/han-
dle/1793/60444/TR1530.pdf.
778

Classification of question answering systems: A survey
S. Sandhini & R. Binu

Computer Science and Engineering Department, Government Engineering College, Palakkad,
Kerala, India
ABSTRACT: Question Answering Systems (QAS) are a unique kind of information

retrieval. In Question Answering (QA) the system retrieves the precise answer to the ques-
tions asked by the user in natural language. QA is multidisciplinary. It involves information
retrieval, natural language processing, linguistics, knowledge representation, databases, soft-
ware engineering, artificial intelligence and so on. This paper classifies QAS based on differ-
ent techniques used in answer ranking and answer extraction. The survey also provides the
main contributions, experimental results, and limitations of various approaches. Finally, we
discuss our perspective on the future direction of QAS.
1 INTRODUCTION
Question Answering (QA) is a particular type of Information Retrieval (IR). The main aim
of Question Answering Systems (QAS) is to retrieve an exact answer to a question in natural
language. The correlation among the QA framework and information retrieval technology, in
IR, the input query containing keywords, and the output consists of a set of documents that
are important to the query asked by a user. QA is not the same as IR in that the user can ask
a question specifically of the system in natural language. The system at that point answers the
question as a concise answer extracted from a source document.
QASs developed in different domains, such as information sources, sorts of questions,
arrangements of answers, and so on; the quantity of such QASs is too huge. The main type
of questions presented by users in natural language are the factoid questions, for example,
“When did the Egyptian revolution occur?”. Also, QA frameworks are arranged into two
main classes, in particular open domain QA systems and closed domain QA systems.
This survey discusses the different methodologies used for answer extraction and the mer-
its and demerits of each. In this paper, Section 2 describes the overview of QAS and different
criteria in support of classifying the large number of QASs available. Section 3 discusses
various future research directions. Section 4 is the conclusion.
2 QUESTION ANSWERING SYSTEMS (QASs)
Figure 1 shows the architecture of a QA framework. The system comprises of three different

modules, for example, question processing, document processing, and answer processing.
The “query processing module” deals with question classification, information retrieval is
done in the “document processing module”, and the answer extraction process is dealt with
in the “answer processing module”.
The question processing module performs question analysis, question classification, and
answer type classification. Question analysis is referred to as the question focus. The question
focus is defined as the word or series of words which shows what information is requested in
the question (Harabagiu et al., 2000). Keeping in mind that the end goal is to answer a ques-
tion correctly, we need to know the type of question requested. Answer type classification
779

Figure 1. QA system architecture.
is related to question classification. In document processing, it retrieves a set of rank docu-

ments that are relevant to the question. It is referred as a paragraph indexing module. The
last phase in the QA design, the answer processing module, is in charge of determining,
retrieving and validating the answers from the list of documents retrieved during the docu-
ment processing module (Allam & Haggag, 2012).
The present search engine or information retrieval system only does the retrieval of the
document, i.e. given a few keywords and to answer some relevant information associated
with the given keyword. Information retrieval does not retrieve the answer. The user wants
an exact response to a question (Hirschman & Gaizauskas, 2001).
The criteria for classifying QAS are the types of techniques used in answer ranking and
also in retrieving answers.
2.1 Classification of QAS based on answer ranking

Here the classification is based on the techniques used in answer ranking such as occurrence
frequency of candidate answers, relevance between information source and question and
similarity between question and answer. Mendes et al. (2014) proposed a method depending
on candidate answers occurrence frequency attention about recognition of logical relation-
ship which includes comparability relationship, consideration relationship and incomplete
connection between candidate answers based on the fact that logical relationship decide the
results of answer arranging. The approach recommended by Toba et al. (2014) and Figueroa
and Neumann (2014) depends on the similarity among question and answer gives more com-
parable is question to answer, greater dependable is the proper response of question. Liu
et al. (2014) propose a methodology depends on relevance among data source and question
gives extra applicable is question to data source, extra valid is the right response retrieved
from the data source.
Mendes et al. (2014) proposes a method to select the answer based on semantic relations.
The fundamental answer selection approach depends on the frequency. It comprises three
steps. First, the candidate answers are normalized to a canonical format, for example date
answers are set to the form D01 M01 Y2015. Second, each two answers are compared. If they
are equivalent, then they are accepted to be the same entity and the score of each answer is
expanded by one. Here, the score of each answer is its frequency of occurrence. Finally, the
relations that exist between each pair of candidate answers are considered. Once more, each
pair of candidate answers is analyzed to distinguish if a relation exists. The answer’s scores
are refreshed, depending upon their relations with different candidates of the event of every
candidate answer: the most frequent candidate answer is picked as the relevant answer.
Toba et al. (2014), propose a method to attempt to show every individual question answer
combine in an unexpected way will move toward to any other maximum which isn’t feasible.
In this, they propose a system which initially considers the question contents to manage the
correct model selection to be utilized for characterization of answer quality, before accom-
plishing the appropriate evaluation of the answer quality.
780

Liu et al. (2014) propose a novel approach recommending essential and complementary
Q&A documents to knowledge groups of QA web sites. The original ideas of this approach
are as follows: it produces community topic profiles by considering QA gathering factors,
such as ‘community member’, the push ranking of QAs and the gathering time of QAs
from the sincerely collected QA facts on particular topics. And it also finds the ranking of
QAs based on question similarity and answer relationship. It proposes a QA-based integral
approach and topic based correlative way to propose complementary QA documents.
2.2 Classification of QAS based on answer extraction

This classifies QAS based on techniques used for retrieving answers. The strategy of answer
extraction depends on text patterns, named entity, and similarity computing between sen-
tences. Wu et al. (2015) propose a technique to find the appropriate answer to the question
which is most extreme similar to question in Q&A dataset is returned in view of sentence
similarity. Ravichandran and Hovy (2002) propose a model for discovering answers by misus-
ing surface text data utilizing manually developed surface patterns. Xu et al. (2003) developed
a method to improve the poor recall of the manual hand-making patterns. Liu et al. (2015)
propose a method based on similarity computing between sentences. Peng et al. (2005) intro-
duced a way to deal with capturing long-distance dependencies through utilizing linguistic
systems to upgrade patterns. Rather than exploiting surface text data utilizing patterns, Lee
et al. (2005) utilized the named-substance way to deal with finding an answer.
Ravichandran and Hovy (2002) use patterns to find the answer to a question based on
the following steps. First, determine the type of the question. Then the question term is
recognized. The third step is to create a query from the question term and perform informa-
tion retrieval. Then segment the documents into different sentences. After that, replace the
question term in each sentence by the question tag. And a pattern table is created for the
particular question type. Finally, sort these answers based on the pattern’s score and return
the top answers.
Xu et al. (2003) introduce definitional QA. This approach combines complementary tech-
nologies, including information retrieval and various linguistic and extraction for analyzing
text. First, the question type is identified, i.e. whether a question is a who or a what question.
Second, information retrieval pulled documents about the question focus from the TREC
corpus. Third, heuristics were applied to the sentences in the retrieved documents to decide
whether they say the question target. Fourth, kernel facts that specify the question target
were extracted from sentences by a variety of semantic preparing and information extraction
tools. Fifth, all kernel facts were positioned by their type and their comparability to the pro-
file of the question. Finally, heuristics were applied to identify repetitive kernel facts.
Lee et al. (2005) developed a hierarchical classifier that classifies questions into fine-grained
classes. For classifying Chinese questions, they presented a machine learning approach (SVM)
and a knowledge-based approach.
2.3 Contributions, experiments and limitations

Mendes et al. (2014)
In this research, the most frequent candidate answer is selected without any other processing.
The selection is mainly based on the relation between the answers. If all candidate answers
have the same frequency, the system probably choose randomly among all answers. Results
show an accuracy of 33.89% for the 438 questions.
Toba et al. (2014)

The framework provides predicting quality of answers in the community QA. The primary
objective of this method is to identify the best machine learning algorithm in terms of feature
selection. The experiment shows that the best machine learning model that fits this approach
is logistic regression.
781

Liu et al. (2014)
This method is to solve the issue of information overloading in community QA web sites.
The experimental results show that the performance of the recommendation approaches. In
this, it considers community topic profiles with QA collection elements, and complementary
scores of QAs, in step with forms better than traditional group primarily based recommenda-
tion methods.
Ravichandran and Hovy (2002)

In this, the technique takes in online data patterns utilizing a few seed questions and answer
anchors, without requiring human comment. A TREC10 question data set is used for the
two sets of experiments. In the first, the TREC corpus is applied as the data source utilizing
an IR factor in their QA framework. Within the second test, the internet is applied as the
data source utilizing the AltaVista web engine to carry out IR. The demerits of this method
are the system performs only certain types of questions. So, in large definitional questions it
performs badly, as the patterns did not deal with lengthy-distance dependencies.
Xu et al. (2003)
This uses various approaches, consisting of data retrieval and distinct linguistic and extrac-
tion tools like parsing, name finding, co-reference resolution, extraction of relation, proposi-
tion and established patterns, so it adopts a hybrid approach. Here they performed three runs
utilizing the F-metric for evaluation. In the main run, BBN2003A, the web was not utilized
as part of the answer finding. In the second run, BBN2003B, answers for factoid questions.
Finally, BBN2003C was the same as BBN2003B except that if the answer for a factoid ques-
tion was discovered various times in the corpus, its score was supported. The performance
for BBN2003A, BBN2003B, BBN2003C runs was 52.1%, 52.0% and 55.5% respectively. The
limitation of this approach, is that the experiments only tested for “what” and “when” ques-
tions. It didn’t consider other factoid questions such as “when” and “where” questions.
Lee et al. (2005)

Their research proposed hybrid architecture for the NTCIR5 and CLQA to answer Chinese
factoid questions. To arrange Chinese questions they displayed a machine learning approach
Table 1. Classification of QA systems.
Classification based on answer ranking
Type of information
Approach Model Key problem source
Based on occurrence Mendes et al. (2014) Logical relationship recognition Open

frequency between candidate answers
Based on similarity Toba et al. (2014) Similarity computing Closed
between questions
Relevance between Liu et al. (2014) Relevance computing Open
information source
and question
Classification based on answer extraction
Approach Model Information source Evaluation
Text patterns Ravichandran and TREC corpus High recall rate

Hovy (2002) Low precision rate
Similarity computing Xu et al. (2003) Yahoo! Answer High precision rate
between sentences Limited answering
Named entity Lee et al. (2005) TREC10 corpus High accuracy
782

(SVM) and a knowledge-based approach (InfoMap). Various approaches are adopted and
integrated including question focus, coarse-fine taxonomy and SVM machine learning. The
method accomplished overall precision, for correct answers and for correct unsupported
answers of 37.5% and 44.5% respectively. Likewise, utilizing InfoMap gave a precision of
88% for the question classification module and for utilizing support vector machine of 73.5%.
3 CONCLUSION AND FUTURE SCOPE
In this survey, we classified QASs on the basis of criteria such as answer ranking and answer
extraction. The occurrence frequency of candidate answers, similarity between question and
answer, relevance between information source and question functions are taken from the
answer ranking method. And, for answer extraction text patterns, named entity and similar-
ity computing between sentences features are taken. In future, valid answer extraction strate-
gies are needed. For open domain QA systems, it returns fake candidate answers, noisy data,
and imprecise candidate answers which influence the final answer.
REFERENCES
Allam, A.M.N. & Haggag, M.H. (2012). The question answering systems: A survey. International Jour-
nal of Research and Reviews in Information Sciences (IJRRIS), 2(3).
Figueroa, A. & Neumann, G. (2014). Category-specific models for ranking effective paraphrases in
community question answering. Expert Systems with Applications, 41(10), 4730–4742.
Gupta, V. & Lehal, G.S. (2009). A survey of text mining techniques and applications. Journal of Emerg-
ing Technologies in Web Intelligence, 1(1), 60–76.
Hao, T. & Agichtein, E. (2012). Finding similar questions in collaborative question answering archives:
toward bootstrapping-based equivalent pattern learning. Information Retrieval, 15(3–4), 332–353.
Harabagiu, S.M., Moldovan, D.I., Pasca, M., Mihalcea, R., Surdeanu, M., Bunescu, R.C. &
Morarescu, P. (2000). FALCON: Boosting knowledge for answer engines. In TREC. 9 (pp. 479–488).
Hirschman, L. & Gaizauskas, R. (2001). Natural language question answering: the view from here.
Natural Language Engineering, 7(4), 275–300.
Kolomiyets, O. & Moens, M.F. (2011). A survey on question answering technology from an information
retrieval perspective. Information Sciences, 181(24), 5412–5434.
Lampert, A. (2004). A quick introduction to question answering. CSIRO ICT Centre.
Lee, C., Shih, C., Day, M., Tsai, T., Jiang, T., Wu, C., Sung, C. & Hsu, W. (2005). ASQA: Academia
Sinica question answering system for NTCIR-5 CLQA. In Proceedings of NTCIR-5 Workshop
Meeting.
Liu, D.R., Chen, Y.H. & Huang, C.K. (2014). QA document recommendations for communities of
question–answering websites. Knowledge-Based Systems, 57, 146–160.
Liu, D.R., Chen, Y.H., Shen, M. & Lu, P.J. (2015). Complementary QA network analysis for QA
retrieval in social question-answering websites. Journal of the Association for Information Science and
Technology, 66(1), 99–116.
Liu, Y., Yi, X., Chen, R. & Song, Y. (2016). A survey on frameworks and methods of question answer-
ing. In Information Science and Control Engineering (ICISCE), 3rd International Conference
Lopez, V., Uren, V., Sabou, M. & Motta, E. (2011). Is question answering fit for the semantic web?
A survey. Semantic Web, 2(2), 125–155.
Mendes, A.C. & Coheur, L. (2011). An approach to answer selection in question-answering based
on semantic relations. In International Joint Conference on Artificial Intelligence (IJCAI)
(pp. 1852–1857).
Mendes, A.C., & Coheur, L. (2013). When the answer comes into question in question-answering:
Survey and open issues. Natural Language Engineering, 19(1), 1–32.
Peng, F., Weischedel, R., Licuanan, A. & Xu, J. (2005). Combining deep linguistics analysis and surface
pattern learning: A hybrid approach to Chinese definitional question answering. In Proceedings of
the Conference on Human Language Technology and Empirical Methods in Natural Language Process-
ing (pp. 307–314). Association for Computational Linguistics.
783

Ravichandran, D. & Hovy, E. (2002). Learning surface text patterns for a question answering system. In
Proceedings of the 40th Annual Meeting of the Association for Computational Linguistics (pp. 41–47).
Association for Computational Linguistics.
Stoyanchev, S., Song, Y.C., & Lahti, W. (2008). Exact phrases in information retrieval for question
answering. In Coling 2008: Proceedings of the 2nd workshop on Information Retrieval for Question
Answering (pp. 9–16). Association for Computational Linguistics.
Toba, H., Ming, Z.Y., Adriani, M. & Chua, T.S. (2014). Discovering high quality answers in community
question answering archives using a hierarchy of classifiers. Information Sciences 261, 101–115.
Voorhees, E.M., & Buckland, L. (2003). Overview of the TREC 2003 Question Answering Track. In
TREC (pp. 54–68).
Wu, Y., Hori, C., Kashioka, H. & Kawai, H. (2015). Leveraging social Q&A collections for improving
complex question answering. Computer Speech and Language, 29, 1–19.
Xu, J., Licuanan, A. & Weischedel, R.M. (2003). TREC 2003 QA at BBN: Answering definitional ques-
tions. In TREC (pp. 98–106).
Zhang, D. & Lee, W.S. (2003). A web-based question answering system. Massachusetts Institute of
Technology (DSpace@MIT).
Zheng, Z. (2002). AnswerBus question answering system. In Proceedings of the Second International
Conference on Human Language Technology Research (pp. 399–404). Morgan Kaufmann Publishers
Inc.
784

A survey of lexical simplification
K.S. Silpa & M. Irshad

ABSTRACT: Lexical simplification is the process of identifying complex words in a sen-

tence and replacing them with simpler substitutes to make the text more accessible to people.
The possible candidate substitutions are generated in the second stage of a lexical simpli-
fication system, known as the substitution generation stage. In this survey, different meth-
odologies used for generating such substitutions are discussed. The merits and drawbacks
associated with each method are also described.
1 INTRODUCTION
While reading a document, article or any other text resources, we might struggle with unfa-
miliar words. Unfamiliar words can be regarded as complex words and these kinds of words
make a text difficult to understand. Jargon, technical terminology, and so forth, are more
difficult to understand for the general population. If everyone wrote texts in the simplest
form, every reader could understand them easily. But these kinds of documents are very
rare. Complex words are always a barrier to comprehending a text. If there was a system to
identify these complex words and replace them with simpler alternatives, these kind of under-
standability problems could be rectified easily and the reader could get enough information
from the text; otherwise, it may be left unread. So, lexical simplification systems have been
introduced, which increase the readability of a sentence by identifying complicated terms
and replacing them with a simpler substitute for better understanding.
A typical Lexical Simplification (LS) system consists of four stages. The first stage of
any lexical simplification system is Complex Word Identification (CWI) and this is the most
important stage of all stages in the LS pipeline. In this stage, the system must identify the
complex word; the candidate substitutions for the identified complex word are generated
later. But the task is very difficult to implement because there is no correct definition for a
complex word. We don’t know how to define a complex word. The complexity of a word
differs from person to person. Different people have different vocabularies depending, for
example, on the newspapers they read, their interaction manner, and so forth.
Either explicitly or implicitly, every lexical simplification system will identify a complex
word. Among the different complex word identification methodologies, simplifying every-
thing is the simplest method but it is not the most effective because it will consider all the
words as complex.
The second stage of LS is known as Substitution Generation (SG). After correctly identi-
fying a complex word, the next step is to generate a suitable substitute for that word. The sub-
stitutes have a similar meaning to the complex word but should be simpler. This can be done
with or without considering the context. A method based on context is the extraction of a
substitute word from the sentence-aligned parallel corpora of English Wikipedia and Simple
English Wikipedia. The context and word alignment in both corpora are considered in order
to generate the substitution. WordNet, using SG methods, is purely context-independent. All
thesauri-based approaches are like this.
785

Not all of the generated substitutions are correct replacements of a complex word. So,
filtering is required to identify the simplest word. The methods of Substitution Selection (SS)
and Substitution Ranking (SR) are used, respectively, to select and rank the candidate sub-
stitutions, which are obtained from the SG stage. Most SS strategies have used Word Sense
Disambiguation (WSD) techniques to avoid ambiguity. The same word may have different
meanings according to the context. So, by using WSD techniques we can select the correct
substitute. SR is for ranking the remaining candidate substitutions based on their simpli
city. The first-ranked substitute is the simplest substitute and, finally, the complex word is
replaced by this one.
Figure 1 describes the lexical simplification pipeline. Here, the complex sentence ‘The cat
perched on the mat’ is taken as the input and, finally, the complex word perched is replaced
with the simpler term, sat.
Lexical simplification systems were first developed for aphasic readers as an assistive tech-
nology. Aphasia is a disability related to understanding language, often caused by a stroke
or head injury. Those who find understanding a text difficult could consider lexical simpli-
fication as an assistive technology. In the same way, LS is used by people who suffer from
cognitive disabilities such as dyslexia. There are several assistive technology projects in LS,
each of them different from one another, based on the needs of the group they are aiming
to assist. Many people often find it difficult to understand technical terms such as medical
terms, scientific terms, and so on. LS is also a solution for these kinds of problem. Another
group of people who have benefited from LS are second-language learners, people with low
literacy skills, children, non-native speakers, and so forth. So, LS is relevant to a wide variety
of different applications.
There are four stages in the lexical simplification process which have been briefly described.
The simplification rules are generated in the second stage, the SG stage. There are several
Figure 1. Stages in the lexical simplification process.
786

approaches for generating simplification rules. This survey aims to discuss the different meth-
odologies for SG and make comparison between them.
The layout of this paper is as follows: Section 2 describes the various approaches for SG;
Section 3 discusses future work in this area; Section 4 makes concluding remarks in relation
to this work.
2 LEXICAL SIMPLIFICATION
Lexical simplification is the task of making difficult words easier to understand. In the proc-
ess of lexical simplification, SG is the generation of all possible candidate substitution words
for a complex word. Figure 2 shows the different approaches for generating the candidate
substitutions.
2.1 Using word-embedding models

Word-embedding models are used to convert a word into vector format. This mapping is based
on the context of the words. In LS, we are trying to find a simple word bearing a similar meaning
to a complex term. Words with similar meanings will be adjacent to each other when converted
to a vector and can thereby be extracted. It results in some complex and simple word pairs.
Kim et al. (2016) proposed a LS method for replacing complex scientific terms with sim-
pler alternatives. In this method, parallel corpora of scientific articles and Wikipedia are
used to extract word pairs. In order to generate the rule, word-embedding models are used,
such as Word2vec and GloVe. Here, Word2vec is used for learning word vectors and it is used
with a skip-gram architecture. The word-embedding dimension is set to 300, and the context
window size is set to 10. After obtaining the word pairs, some filtering methods are applied
to obtain the most appropriate substitutions, which are regarded as a kind of SS and SR
approach. This includes calculating the cosine similarity between the words in a rule, elimina-
tion of word pairs with different parts of speech, and so forth.
Word2vec is a very effective tool for vector formation and fast training is also possible with
it. The methodology of Kim et al. (2016) also provides a higher precision and is more efficient
than constructing a co-occurrence matrix as proposed in Biran et al. (2011). However, the
limitation of Word2vec is that it will map to antonyms instead of synonyms; in other words,
instead of complex word to simpler synonym mapping, it results in complex word to simpler
antonym mapping. This is because both antonyms and synonyms of a particular word may
have the same context and so they appear adjacent in the vector representation. So, from the
simplification rules, the antonyms must be filtered out.
Paetzold and Specia (2016) proposed a context-aware, word-embedding model for SG that
uses a sense label assigned to each of the candidate words in the training corpus. Each sense of
a particular word has a different numerical vector. The sense labels are similar to the Part-Of-
Speech (POS) tag. But this kind of model will identify the inflected form of nouns and verbs as
a different tag. To rectify this problem, all tags related to nouns, verbs, adjectives and adverbs
are marked as N, V, J and R, respectively. Then the POS-tag annotated corpus is trained with
models like GloVe. The Paetzold and Specia (2016) methodology uses a candidate generation
Figure 2. Classification of methodologies used for generating simplification rules.
787

algorithm, which helps to filter out the false and ungrammatical candidates. Thus, problems
associated with ambiguity can be reduced. Traditional word-embedding models cannot accom-
modate the different senses of a particular word; in other words, for every sense of a word, it
has only one numerical vector. So, this model is helpful in eliminating the limitations of tra-
ditional word-embedding models. The training corpus is huge, so giving a sense label to each
of them is tedious work. This difficulty is a major problem associated with this methodology.
Paetzold and Specia (2017) proposed a methodology for SG, in which the context-aware
word embeddings are enhanced with lexicon retrofitting. Retrofitting is always applied as
a post-processing task to improve the quality of the vector. In this methodology, the ret-
rofitting is performed with the help of WordNet over lexicon relations such as synonyms.
The final result is dictionaries that contain a word-tag pair and its corresponding synonym.
Faruqui et al. (2014) introduced a graph-based retrofitting algorithm, which is also used in
this methodology. Sometimes, especially with more dimensions, word vectors are much more
helpful for extracting semantic information.
2.2 Using sentence alignment

The methodology proposed by Horn et al. (2014) uses the sentence-aligned corpora of English
Wikipedia (EW) and Simple English Wikipedia (SEW) for rule extraction. The words in
EW are taken as normal words and the words in SEW are taken as simple words. There is a
sentence-aligned simple sentence corresponding to each normal sentence in the corpus. Due
to the same alignment of words in a normal and simple sentence, the candidate pairs can be
easily extracted. Using the tool GIZA++, an alignment is automatically induced between the
words, which helps to identify word pairs.
It is possible to have errors in word alignment and sentence alignment. So, we cannot guar-
antee that all alignments are correct and all extracted word pairs are real. Some filtering tasks
need to be carried out. One of these is to remove the word pairs in which the normal words
already exist in the stoplist. The stoplist is a list which contains only the simplest terms. If the
normal word is in the stoplist, no further simplification is required. Similarly, if the normal
word is a proper noun, then the corresponding word pairs can also be removed. The POS tags
of these two words in a word pair are checked to ensure that they are same.
The main advantage of using the methodology proposed by Horn et al. (2014) is that it is
helpful for generating a large number of simplification rules; in other words, the coverage is
high. Large numbers of simplification rules lead to a simplification where the preservation
of meaning is high. It is necessary to ensure that meaning and grammar is preserved in every
LS. The method of Horn et al. performs well in the case of preservation of meaning. For
any kind of sentence-aligned approach, there is a problem of one to two mappings of words
between the SEW and EW: this will generate false word pairs that need to be filtered out.
Paetzold and Specia (2017) proposed two different methodologies for SG. One method-
ology is based on sentence alignment. It uses the Newsela corpus for extracting candidate
substitutions. The Newsela corpus contains professionally simplified articles. Sentence align-
ment is then generated on all the pairs of versions of the Newsela articles. The paragraph
and sentence-alignment algorithms proposed by Paetzold and Specia (2016) are used in this
approach. Term Frequency–Inverse Document Frequency (TF-IDF) similarity is used to
identify the correct paragraph and sentence alignment. After producing such alignments, the
methodology proposed by Horn et al. (2014) is used to extract the word pairs.
The SG in Paetzold and Specia (2017) produces fewer grammatical and meaning errors
and achieves the highest precision and F1 scores. But the disadvantage is that this methodol-
ogy may cause a ranking error. Ranking errors are replacements which do not simplify the
selected sentence.
2.3 Using thesauri

De Belder and Moens (2010) proposed a text-simplification system for children, in which
both syntactic and lexical simplifications are employed and simplified news articles and
788

Wikipedia articles are used. In the lexical simplification task, WordNet is used, along with
the Latent Words Language Model (LWLM) for extracting the synonyms. This methodology
selects words as candidate substitutions which are both identified as a synonym in WordNet
and selected by LWLM. Kucera–Francis frequency is used to score the candidates to obtain
the simplest word. The word with highest Kucera–Francis frequency is considered to be the
simplest synonym among the candidate synonyms.
The main disadvantage of using the methodology of De Belder and Moens (2010) is that
it doesn’t replace the most complex terms. This is because it is very difficult to identify a
synonym for these kinds of complex terms. An alternative method is to replace the complex
term with a brief explanation of that term, but this is not considered to be lexical simplifica-
tion. As mentioned before, when using the methodology of Kim et al. (2016), antonyms can
also be wrongly identified as simpler synonyms. These kinds of mistakes are not made in this
approach. This is because WordNet contains a correct distinction between synonyms and
antonyms.
There are several simplification systems for the English language. Bott et al. (2012) pro-
posed a lexical simplification system for the Spanish language. In this system, on-line dic-
tionaries are used for extracting the synonyms. There are no parallel corpora for the Spanish
language, so parallel corpus-based methodologies are not relevant. Like WordNet in English,
the Spanish language has Euro WordNet and the Spanish OpenThesaurus. To find a suitable
synonym for a word, it checks all the possible alternatives of the given word in the Spanish
OpenThesaurus. The selected words are then converted into word vectors. Word frequency
and word length-based approaches are used to obtain simpler synonyms.
The main advantage of this methodology is that it notably preserves the meaning of text
and it can be applied in any language with the help of resources available in that language.
However, problems arise with words that are not in the dictionary. Many words may not
appear in the dictionary; in these cases, the number of alternatives is minimized. So, there is
a need to apply a word sense disambiguation technique for better performance.
Table 1. Overview of the compared approaches and datasets used for lexical simplification.
Proposed approaches
Type of
approach Model Dataset Result
Word- Kim et al. Scientific corpus of nearly 500k f-measure – 0.285

embedding (2016) publications in Public Library of Science Precision – 0.389
model (PLOS) and PubMed Central (PMC),
paired with a general corpus from Wikipedia
Paetzold and Newsela (version 2016-01-29.1) contains Precision – 0.337
Specia 1,911 news articles in their original form, f1 score – 0.256
(2017) as well as up to five simplified versions
Paetzold and Corpus of movie subtitles Precision – 0.118
Specia Recall – 0.161
(2016) f1 score – 0.136
Sentence Horn et al. A sentence-aligned data set of English Precision – 0.761
alignment (2014) Wikipedia sentences and Simple English Accuracy – 0.663
Wikipedia sentences, containing 137,000
aligned sentence pairs
Paetzold and Newsela (version 2016-01-29.1) contains Precision – 0.453
Specia 1,911 news articles in their original form,
(2017) as well as up to five simplified versions
Thesaurus De Belder and Simplified news articles and Accuracy – 0.41
Moens (2010) encyclopedia articles
Bott et al. On-line dictionary and the Web Precision – 0.121
(2012)
789

3 FURTHER SUGGESTIONS
The degree of simplicity and the preservation of meaning and grammar are the main focuses of
any lexical simplification system. If more substitutions are produced, then the degree of mean-
ing preservation will be reduced and vice-versa. This problem is difficult to solve. The perform-
ance of SG can be improved by increasing the number of thesauri used. A suitable method for
complex word identification is still a problem in LS, because there is no correct definition for a
complex word and the complexity varies according to each individual’s vocabulary. There are
some domain-specific LSs available; it can also be extended to other domains.
4 CONCLUSION
LS systems are used to make a text more accessible. One of the main tasks in LS is substi-
tution generation, which is used to generate the possible complex and simple word pairs.
Several methods are used for generating these pairs. The methods include word-embedding
models, thesauri and sentence alignment. Among these methods, word embedding is the
most widely used. The positives and negatives of the different methods are discussed in this
paper. SG methods have an important role in the preservation of grammar and meaning of
the generated sentence.
REFERENCES
Adel, H. & Schütze, H. (2014). Using mined coreference chains as a resource for a semantic task. In Pro-
ceedings of the 2014 Conference on Empirical Methods in Natural Language Processing (EMNLP),
25–29 October 2014, Doha, Qatar (pp. 1447–1452). Stroudsburg, PA: Association for Computational
Linguistics.
Biran, O., Brody, S. & Elhadad, N. (2011). Putting it simply: A context-aware approach to lexical simpli-
fication. In Proceedings of the 49th Annual Meeting of the Association for Computational Linguistics:
Human Language Technologies: Short papers (Vol. 2, pp. 496–501). Stroudsburg, PA: Association for
Bott, S., Rello, L., Drndarevic, B. & Saggion, H. (2012). Can Spanish be simpler? LexSiS: Lexical Sim-
plification for Spanish. In Proceedings of COLING 2012: 24th International Conference on Computa-
tional Linguistics: Technical Papers (pp. 357–374).
Carroll, J., Minnen, G., Canning, Y., Devlin, S. & Tait, J. (1998). Practical simplification of English
newspaper text to assist aphasic readers. In Proceedings of the AAAI-98 Workshop on Integrating
Artificial Intelligence and Assistive Technology (pp. 7–10).
Coster, W. & Kauchak, D. (2011). Learning to simplify sentences using Wikipedia. In Proceedings of the
49th Annual Meeting of the Association for Computational Linguistics, 24 June 2011, Portland, Oregon
(pp. 1–9). Stroudsburg, PA: Association for Computational Linguistics.
Daelemans, W., Höthker, A. & Erik Tjong Kim Sang. (2004). Automatic sentence simplification for
subtitling in Dutch and English. In Proceedings of the 4th International Conference on Language
Resources and Evaluation, Lisbon, Portugal (pp. 1045–1048).
De Belder, J. & Moens, M.F. (2010). Text simplification for children. In Proceedings of the SIGIR Work-
shop on Accessible Search Systems (pp. 19–26).
Deléger, L. & Zweigenbaum, P. (2009). Extracting lay paraphrases of specialized expressions from
monolingual comparable medical corpora. In Proceedings of the 2nd Workshop on Building and Using
Comparable Corpora: From Parallel to Non-Parallel Corpora (pp. 2–10). Stroudsburg, PA: Associa-
tion for Computational Linguistics.
Devlin, S. & Unthank, G. (2006). Helping aphasic people process online information. In Proceedings of
the 8th International ACM SIGACCESS Conference on Computers and Accessibility (pp. 225–226).
New York, NY: Association for Computing Machinery.
Elhadad, N. & Sutaria, K. (2007). Mining a lexicon of technical terms and lay equivalents. In Proceed-
ings of the Workshop on BioNLP 2007: Biological, Translational, and Clinical Language Processing,
29 June 2007, Prague, Czech Republic (pp. 49–56). Stroudsburg, PA: Association for Computational
Linguistics.
790

Faruqui, M., Dodge, J., Jauhar, S.K., Dyer, C., Hovy, E. & Smith, N.A. (2014). Retrofitting word vectors
to semantic lexicons. arXiv:1411.4166.
Horn, C., Manduca, C. & Kauchak, D. (2014). Learning a lexical simplifier using Wikipedia. In
Proceedings of the 52nd Annual Meeting of the Association for Computational Linguistics (Vol. 2,
pp. 458–463). Stroudsburg, PA: Association for Computational Linguistics.
Huenerfauth, M., Feng, L. & Elhadad, N. (2009). Comparing evaluation techniques for text readabil-
ity software for adults with intellectual disabilities. In Proceedings of the 11th International ACM
SIGACCESS Conference on Computers and Accessibility (pp. 3–10). New York, NY: Association for
Computing Machinery.
Jonnalagadda, S., Tari, L., Hakenberg, J., Baral, C. & Gonzalez, G. (2009). Towards effective sentence
simplification for automatic processing of biomedical text. In Proceedings of Human Language Tech-
nologies: The 2009 Annual Conference of the North American Chapter of the Association for Compu-
tational Linguistics, Companion Volume: Short Papers (pp. 177–180). Stroudsburg, PA: Association
for Computational Linguistics.
Kajiwara, T., Matsumoto, H. & Yamamoto, K. (2013). Selecting proper lexical paraphrase for children.
In Proceedings of the Twenty-Fifth Conference on Computational Linguistics and Speech Processing
(ROCLING 2013) (pp. 59–73).
Kim, Y.S., Hullman, J. & Adar, E. (2015). DeScipher: A text simplification tool for science journalism.
Ann Arbor, 1001, 48109.
Kim, Y.S., Hullman, J., Burgess, M. & Adar, E. (2016). SimpleScience: Lexical simplification of scien-
tific terminology. In Proceedings of the 2016 Conference on Empirical Methods in Natural Language
Processing, 1–5 November 2016, Austin, Texas (pp. 1066–1071). Stroudsburg, PA: Association for
Och, F.J. & Ney, H. (2000). Improved statistical alignment models. In Proceedings of the 38th Annual
Meeting of Association for Computational Linguistics (pp. 440–447). Stroudsburg, PA: Association
for Computational Linguistics.
Paetzold, G. & Specia, L. (2015). LEXenstein: A framework for lexical simplification. In Proceedings
of ACL-IJCNLP 2015 System Demonstrations, 26–31 July 2015, Beijing, China (pp. 85–90). Strouds-
burg, PA: Association for Computational Linguistics.
Paetzold, G.H. & Specia, L. (2016). Unsupervised lexical simplification for non-native speakers. In
Proceedings of the Thirtieth AAAI Conference on Artificial Intelligence, 12–17 February 2016, Phoe-
nix, Arizona (pp. 3761–3767). Palo Alto, CA: AAAI Press.
Paetzold, G.H. & Specia, L. (2017). Lexical simplification with neural ranking. In Proceedings of the
15th Conference of the European Chapter of the Association for Computational Linguistics, 3–7 April
2017, Valencia, Spain: Short Papers (Vol. 2, pp. 34–40). Gothenburg, Sweden: Association for Com-
Shardlow, M. (2013). A comparison of techniques to automatically identify complex words. In Proceed-
ings of the ACL Student Research Workshop, 4–9 August 2013, Sofia, Bulgaria (pp. 103–109). Strouds-
burg, PA: Association for Computational Linguistics.
Thomas, S.R. & Anderson, S. (2012). WordNet-based lexical simplification of a document. In Proceed-
ings of the 11th Conference on Natural Language Processing (KONVENS 2012), 19 September 2012,
Vienna, Austria (pp. 80–88).
791

Fast and efficient kernel machines using random kitchen sink and
ensemble methods
P. Melitt Akhil & K. Rahamathulla

Department of Computer Science and Engineering, Government Engineering College, Thrissur,
Kerala, India
ABSTRACT: The introduction of the information era has affected all areas of computer
science. This has also affected the approach taken by machine learning researchers. Com-
pared to the early days, researchers now use a generic approach to selecting a machine
learning model. As the generic model doesn’t contain much domain knowledge, it has been
compensated for by a huge dataset. This makes the model optimization more complex and
time consuming. We can replace this overhead with randomization instead of optimization.
There are many existing methods which apply randomization in machine learning models.
But all of these methods compromise on accuracy. This paper shows an efficient way to use
randomization along with ensemble methods without a decrease in efficiency.
1 INTRODUCTION
As the entire world is tending toward the Internet of Things and Software Defined Anything,
the volume and velocity of data is exploding. This has led to certain changes in a researcher’s
approach in machine learning area too. In the early days of machine learning, researchers
used complex models. Most of the domain knowledge will be embedded in these models. And
the models will be application dependent. There won’t have been huge datasets available for
optimization purposes at that time. But now, with the help of world wide web, the Internet of
Things and other information era facilities, there is a possibility of getting huge sets of data.
This huge dataset will contain most of the domain knowledge for a particular application.
This has made researchers follow a generic approach to model selection.
Because of the availability of huge datasets, there is no need to embed the complete domain
knowledge into the model. This reduces the development time of models. The generic models
can be reused for similar applications which, in turn, improve a particular type of model as
it has been used for many projects. As the training data contains more accurate features, the
models optimized using this data will be more accurate.
But the huge size of datasets has increased the training time exponentially. If the input
dataset contains N data points, then the training time of a kernel machine will be N2. Also, it
is difficult for storing kernel matrices because of its huge size, which is N × N.
To overcome this overhead, there are many approaches, such as dimension reduction, con-
verting a gram matrix from dense to sparse, decomposition of a gram matrix, random pro-
jection and sampling. But all of these methods lack accuracy because when the quantity of
training data decreases, features obtained from the dataset also decrease.
In order to have accurate and fast training models, methods which decrease training time
and increase accuracy, need to be integrated. This paper proposes a method which uses a
sub-sampling method and an ensemble method together to obtain a fast and efficient kernel
machine.
Section 2 introduces the Random Kitchen Sink (RKS) and ensemble methods. In Section 3
the new methodology is explained. Section 4 analyses the evaluation criteria for the new
model.
793

2 RELATED WORKS
Random Kitchen Sink (RKS) (see 2.2) and other random approaches (see 2.1) are explained
in detail in this section. Different types of ensemble methods (see 2.3) are also studied.
2.1 Randomization methods

SVM uses Quadratic Programming (QP) methods to solve the model. But the QP which
comes from SVM cannot be solved easily. So, methods like chunking (Vapnik) and decom-
position are used widely (Platt, 1999). Chunking which reduced the dimension of the matrix
breaks down the large QP into smaller ones. But even chunking cannot handle large data
set problems. This motivated the decomposition method which uses a fixed-size matrix for
every QP sub-problem. Sequential Minimal Optimization (SMO) (Platt, 1999) also follows
the fixed matrix size approach to decomposition. But it chooses a greedy approach in solving
optimization problem. SMO solves the smallest optimization problem at each iteration. This
avoids numerical QP optimization. Thus, kernel evaluation is the only major factor that con-
tributes to computation time in SMO. Kernel evaluation time can be reduced by increasing
the sparseness of the input data.
In order to get a sparse gram matrix many approximation methods were used. One method
is Singular Value Decomposition (SVD) which approximates gram matrices. But, due to the
large size of gram matrices, finding SVD is a time complex task. This can be avoided by ran-
domly sampling the gram matrix. Rows and then columns are chosen independently accord-
ing to normal distribution (Frieze et al., 2004). After that it is easy to find the SVD of the
resultant matrix as the resultant matrix is of a smaller size.
Also there are methods which formulate the relationship between SVM and a Minimum
Enclosing Ball (MEB) problem (Tsang et al., 2005). This leads to a more simple and efficient
way of approximating large data sets.
Willims and Seeger (2001) recommended the Nyström method. The Nyström method con-
tains a matrix C of size n × c, which contains randomly chosen columns c. It also has a matrix
W, which contains the intersections of randomly chosen columns and their corresponding
rows. This method is modified by allowing the column sample to be formed using arbitrary
sampling probabilities (Drineas & Mahoney, 2005).
One of the sampling techniques uses a three-tier approach for speeding Kernel Principal
Component Analysis (Achlioptas et al., 2001). At the first stage, the gram matrix is sam-
pled with probability factor s. This matrix is then quantized with the maximum value in the
matrix. After this, a randomized rounding is conducted for the kernel coefficients. This helps
for a fast batch approximation of the kernel function.
Random projection is more efficient than normal sampling techniques. It utilizes mapping
elements to a d-dimensional subspace where d is much smaller than the actual dimension
(Blum, 2006). The Anchor method is a random projection method (Wang et al., 2017), where
randomization is embedded in the machine learning methods. The features are extracted, not
for each data point, but for a group of points called anchors. Anchor granularity is decided
randomly using geometric distribution. This approach removes over-fitting and noise in the
model.
2.2 Random Kitchen Sink (RKS)

By mapping the data points from the original low dimension space to a higher infinite dimen-
sional space we can easily separate the data points according to various clusters. This is the
main objective of the kernel trick. But this is time consuming and costly. In RKS, the com-
plexity is reduced by mapping to a low dimensional Euclidean inner product space. This uses
a randomized feature map z: Rd–>RD, where z is low dimensional.
Some of the randomized maps are sinusoids randomly drawn from the Fourier transform
of the kernel function, and randomly shifted grids which partitioned the input space (Rahimi
& Recht, 2007).
794

RKS is a generalized randomization method for any type of machine learning model.
Every optimization problem can be viewed as a function fitting problem. Each function fit-
ting problem tries to reduce the empirical risk equation. The generalized empirical risk func-
tion contains parameters, ω, and weights, α, over K features. It uses a minimization process
for both weights and parameters.
But RKS randomizes the parameters, ω, and minimizes the weights, α, using batch opti-
mization (Rahimi & Recht, 2008b). Randomization instead of optimization decreases the
computational complexity of training (Rahimi & Recht, 2008a).
There are also methods which use Hadamard and diagonal matrices instead of Gaussian
matrices in RKS (Quoc et al., 2013).
Some of the applications of RKS are retrieval of bio-geo-physical parameters (Laparra
et al., 2015), hyper spectral image classification (Nikhila, 2015; Nikhila et al., 2015), Tamil
document classification (Sanjanasri & Anand, 2015), offline power disturbance signal classi-
fication (Aneesh et al., 2015) and detection of mitotic nuclei in breast histopathology images
(Beevi et al., 2016).
2.3 Ensemble method

Ensemble classifiers can create a strong learner with the help of weak learners. It will have a
base learner. The results of these base learner are combined using different rules. The com-
bining rule can weight the base learners according to the error rate (Xiao & Wang, 2017).
Also, top performing base learners can be found out. At last in order to get a single output
voting system can be implemented where majority voting output of each base learner will be
counted as final output.
It is hard to predict multi-regime time series datasets using a single model. But using
ensemble methods we can reduce the error rate. The random forest ensemble method uses
bagging and random subspace techniques for achieving diversity (Lin et al., 2017).
Liver tumor detection (Huang et al., 2014) and predicting real exhaust gas temperatures
for turbofan engines (Lin et al., 2017) are some of the examples where different ensemble
methods are used.
3 PROPOSED METHOD
3.1 Mathematical model

Every randomization method comes with an expected error factor. And, in every case, this
error factor decreases the accuracy of the model. To overcome this compromise in accuracy,
we have integrated the ensemble method with randomization. The randomization method
selected for the classification is RKS as this method has less of an error rate.
In order to create the voting system for the entire model first, the voting share of each
learner needs to be calculated. It depends on the error rate of each learner (Equation 1).
1 m
Ej = ∑ I(x i ) − y i
m i=1
(1)
Then the voting share can be calculated with proportions of error rate distributed between
the individual learner and entire learners using Equation 2.
∑
n
Ei
Vj = i =1
(2)
Ej
The training of base learners is done using the kernel which was built by RKS. The steps
of the proposed method are described in algorithm 3.1.
795

3.2 System design
First, the entire data set is partitioned into multiple overlapping small datasets using prob-
ability distribution. Then using RKS, different kernel spaces are created in parallel. Using
those kernel spaces, base learners are trained. The output of each base learners is fed into
the voting system (Figure 1). The voting system gives a weighted voting share to each base
learner. It then calculates the final output using majority voting share.
Figure 1. Model representation.
796

4 EVALUATION CRITERIA
The error rate of the proposed method depends only on the error rate of each base learner
(Equation 3). The randomization process done during the RKS affects this error. But because
a weighted voting share system is used, the error rate of each learner will be constant. So, the
ultimate error rate of the entire model will be the smallest error rated learner.
E = max E j × V j (3)
j =1…n
This error is less than normal randomization methods which use only one learner. Also,
in the case of time complexity, our model is faster than the classical kernel machine. It takes
only O (dlogd) time, where d is the size of the dataset for each learner which is obtained
by partitioning the entire input space. Here, d << n. As the size of data set for each kernel
machine is smaller, it takes up much less space to store the gram matrix (d × d). Thus, spatial
complexity also decreases.
5 CONCLUSIONS
We proposed a method which combines advantages from both randomization and ensemble
methods. Without losing accuracy, the kernel is trained faster than a normal kernel. The algo-
rithm splits the entire dataset and, for each piece, a base learner is trained using RKS. The
weighted voting share method is integrated with the system in order to decrease the error rate.
The time and space complexity of the problem also decreases due to the randomization process.
REFERENCES
Achlioptas, D., McSherry, F. & Scholkopf, B. (2001). Sampling techniques for kernel methods. In
Advances in Neural Information Processing Systems (335–342), 14.
Aneesh, C., Hisham, P.M., Sachin, K.S., Maya, P. & Soman, K.P. (2015). Variance based offline power
disturbance signal classification using support vector machine and random kitchen sink. Procedia
Technology, 21(21), 163–170.
Beevi, K.S., Madhu, S.N. & Bindu, G.R. (2016). Detection of mitotic nuclei in breast histopathology
images using localized acm and random kitchen sink based classifier. 38th Annual International Con-
ference of the IEEE Engineering in Medicine and Biology Society (EMBC), 2435–2439.
Blum, A. (2006). Random projection, margins, kernels, and feature-selection. SLSFS ’05 Proceedings of
the 2005 International Conference on Subspace, Latent Structure and Feature Selection, 52–68.
Drineas, P. & Mahoney, M.W. (2005). On the Nyström method for approximating a gram matrix for
improved kernel-based learning. Journal of Machine Learning Research, 2153–2175.
Frieze, A., Kannan, R. & Vempala, S. (2004). Fast Monte-Carlo algorithms for finding low-rank
approximations. Journal of the ACM (JACM) 51, 1025–1041.
Huang, W., Yang, Y., Lin, Z., Huang, G.-B., Zhou, J., Duan, Y. & Xion, W. (2014). Random feature
subspace ensemble based extreme learning machine for liver tumor detection and segmentation. 36th
Annual International Conference of the IEEE Engineering in Medicine and Biology Society, 4675–4678.
Laparra, V., Gonzalez, D.M., Tuia, D. & Camps-Valls, G. (2015). Large-scale random features for kernel
regression. IEEE International Geoscience and Remote Sensing Symposium (IGARSS), 17–20.
Lin, L., Wang, F., Xie, X. & Zhong, S. (2017). Random forests-based extreme learning machine ensem-
ble for multi-regime time series prediction. Expert Systems with Applications, 83, 164–176.
Nikhila, H., Sowmya, V. & Soman, K.P. (2015). Comparative analysis of scattering and random features
in hyperspectral image classification. Second International Symposium on Computer Vision and the
Internet (VisionNet15) 58, 307–314.
Nikhila, H. (2015). Hyperspectral image classification using random kitchen sink and regularized least
squares. IEEE International Conference on Communication and Signal Processing (ICCSP).
Platt, J.C. (1999). Using analytic QP and sparseness to speed training of support vector machines. In
Proceedings of the 1998 Conference on Advances in Neural Information (pp. 557–563).
797

Quoc, L., Sarlos, T. & Smola, A. (2013). Fastfood approximating kernel expansions in loglinear time.
30th International Conference on Machine Learning (ICML).
Rahimi, A. & Recht, B. (2007). Random features for large-scale kernel machines. In Advances in Neu-
ral Information Processing Systems, 20 (NIPS) (pp. 1177–1184), Vancouver, British Columbia,
Canada.
Rahimi, A. & Recht, B. (2008a). Uniform approximation of functions with random bases. 46th Annual
Allerton Conference on Communication, Control, and Computing, 555–561.
Rahimi, A. & Recht, B. (2008b). Weighted sums of random kitchen sinks: Replacing minimization with
randomization in learning. In Advances in Neural Information Processing Systems, 21 (NIPS) (pp.
1313–1320), Vancouver, British Columbia, Canada.
Sanjanasri, J.P. & Anand, K.M. (2015). A computational framework for Tamil document classification
using random kitchen sink. International Conference on Advances in Computing, Communications and
Informatics (ICACCI), 1571–1577.
Tsang, I.W., Kwok, J.T. & Cheung, P.-M. (2005). Core vector machines: fast SVM training on very large
data sets. Journal of Machine Learning Research, 363–392.
Wang, S., Aggarwal, C. & Liu, H. (2017). Randomized feature engineering as a fast and accurate alter-
native to kernel methods. KDD ’17 Proceedings of the 23rd ACM SIGKDD International Conference
on Knowledge Discovery and Data Mining, 485–494.
Xiao, Q. & Wang, Z. (2017). Ensemble classification based on random linear base classifiers. IEEE
International Conference on Acoustics, Speech and Signal Processing (ICASSP), 2706–2710.
798

HOG feature-based recognition for Malayalam handwritten

characters
E.P. Anjali & Ajay James

Department of Computer Science and Engineering, Government Engineering College,
Saravanan Chandran
Department of Computer Science and Engineering, National Institute of Technology,
Durgapur, West Bengal, India
ABSTRACT: Optical Character Recognition (OCR) converts images containing handwrit-

ten or printed characters into an editable format. OCR systems have wide applications such
as processing bank cheques, or document conversion of legal papers. One major concern in
developing a character recognition system is the selection of efficient features. Major chal-
lenges in Malayalam handwritten character recognition are varying writing styles, the pres-
ence of compound characters and of similarly shaped characters. Mixing up old and new
styles of writing adds additional complexity in HCR systems. In this paper, an alternative
that allows the use of histogram of oriented gradients is presented. The proposal consists
of using HOG for feature extraction, then using SVM as the classifier, Malayalam character
recognition can be carried out with more quality than that obtained in the existing methods.
1 INTRODUCTION
The entire world is now moving towards digitization and is growing fast where printed doc-
uments and manuscripts need preserved in a digital format. Optical Character Recognition
(OCR) is a system by which any document, whether it is handwritten or printed, can be con-
verted into an editable text format. Thus, it helps when searching in documents. Nowadays,
OCR is used by a large number of institutions such as banks, insurance companies, post offices
and book publishers in order to verify the authenticity of the documents, either handwritten
or printed such as cheques, and envelopes. The research and development of OCR is based on
progress in fields such as image processing, pattern recognition and machine learning. There
are two types of OCR systems: offline and online. In an offline character recognition system,
the input image is captured by a scanner and passes through three stages; preprocessing, feature
extraction and classification. Online character recognition systems work in real-time.
The first step in every traditional recognition system is the removal of noise. This is followed
by other preprocessing techniques such as binarization, thinning, resizing. The preprocessed
image is then segmented into lines, words and then characters. Unique features are extracted
from each character and the chosen classifier is trained using these vectors. Compared to
printed documents, handwritten character recognition is more complex since different people
write in different ways and styles. Other challenges in handwritten character recognition are
the variation in the fonts/thickness of letters and the difference in the gaps between the letters.
Moreover, the skewness of handwritten matter will be very different from person to person
(Ryu et al., 2014). To overcome these challenges, researchers are working hard to improve this
system.
This paper has other parts such as insight on previous works in OCR using HOG (Sec. 2),
proposed method (Sec. 3) and conclusion (Sec. 4).
799

2 PREVIOUS WORK
Elleuch et al. (2017) have extracted features using a HOG descriptor. The image of each char-
acter is initially divided into small cells. The HOG of each pixel is then computed using a one-
dimensional mask. They have represented a rectangular HOG using three parameters namely,
“number of cells per block, number of pixels per cell, and number of channels per cell”. The
orientation is taken in-between 0 and 180 degrees. They have chosen nine bins to represent orien-
tation. After computing the histogram, normalization is done by L2-norm. They have compared
the performance of HOG features with Gabor features. Gabor features are directly extracted
from gray scale images. Their system was tested using an IFN/ENIT dataset. The error classifi-
cation rate of the HOG descriptor with SVM was 1.51% and that of Gabor features was 7.16%.
Elleuch et al. (2015) examined Arabic handwritten script recognition using a multi-class
SVM. In the recognition phase, structural features are input to a supervised learning algo-
rithm. SVM is used as the classifier and handwritten Arabic characters are used for testing.
Their experimental study has efficiently showed that they have excellent outcomes if com-
pared to the existing Arabic OCR systems.
Ebrahimzadeh and Jampour (2014) have developed an efficient handwritten digit recogni-
tion system using HOG and SVM. The input image is partitioned into 9 × 9 cells and a histo-
gram is computed. Eighty-one features are used to represent each digit. Digit classification is
performed using linear multi-class SVM. To validate the model, a MNIST handwritten digit
dataset is used. They have achieved 97.25% accuracy. Linear SVM yields better accuracy
compared to polynomial, RBF and sigmoid functions.
Kamble and Hegadi (2015) use Rectangular HOG (RHOG) features to recognize hand-
written Marathi characters. Normalization will bring all characters to the same size. After
extracting RHOG features, the feature vector length is 576. A Sobel mask is used to measure
the gradient values. After calculating the gradient and orientation, bins of histograms are
computed. SVM is used as the classifier. The performance is compared with a Feed forward
artificial neural network. The dataset included 8,000 samples and the neural network-based
classification performed better than SVM.
The handwritten character recognition in Kulkarni’s (2017) model uses the HOG features
for character recognition followed by the center of the mass of the image with SVM algo-
rithm. Otsu’s method is used for segmentation. Once regions of interest are obtained, the
mean of the weighted mean of white pixels is calculated. The center of image is assigned this
value. 9-bit integer values are extracted using the HOG descriptor. For character classifica-
tion, the HASY dataset is used which contains handwritten alpha numeric symbols. The
proposed model is evaluated using SVM and KNN.
Qinyunlong (2008) combines HOG in multiple resolutions with canonical correlation anal-
ysis. A Gaussian pyramid is used to get a multi-resolution HOG. Once we have the gradient
map, HOG features can be extracted from these maps for each resolution. In preprocessing,
Box-Cox transformation is applied. The system is tested with three handwritten databases.
Iamsa-at and Horata (2013) The images are converted into gray scale and after preproc-
essing, the resized images are at 32 × 32 pixels. HOG was computed by applying a one-
dimensional mask. Based on the intensity of the characters, the gradient is calculated. Each
pixel casts a vote for the cell that lies closest to its orientation. L2 norm is used to normalize
the histograms of overlapping blocks. The dataset contains Thai and Bangla handwritten
characters. The performance of the feedforward-backpropagation neural network and the
Extreme Learning Machine (ELM) are compared. Eighty hidden layers were used in the
backpropagation neural network and the activation function is logistic. A sigmoid function is
used to train the ELM. Their experimental study shows that DFBNN outperforms ELM.
Tikader and Puhan (2014) have modified the traditional HOG feature for recognizing
English-Bengali scripts. The input image is not divided into cells. After computing gradients,
instead of splitting the cells, binning operation is applied to the whole image. For classifica-
tion, linear SVM is used. The system performance depends on the number of bins chosen
since they are proportional to each other.
800

Newell and Griffin (2011) proposed multiscale HOG descriptors for robust character recog-
nition. Their work describes two variations of using HOG features. Their first scheme simply
extends the histograms across different scale space. Orientation is calculated using Derivative-
of-Gaussian filters. In the second method, pairs of oriented gradients across different scales
are integrated to generate feature vectors. This is referred to as oriented gradient columns.
3 PROPOSED METHOD
Due to the high variety and complexity of Malayalam handwritten strokes, shapes and con-
cavities, we have selected a HOG feature descriptor which is robust to local displacements, yet
still supply discriminating feature vectors as figuration of the handwritten characters. Bhow-
mik et al. (2014) mention “The main idea behind the HOG descriptors is that the local object
appearance and shape within an image can be described by the distribution of intensity or
edge directions.” Then, obtained feature vectors are fed into an SVM classifier to create the
classification. The system works in three steps:
• Preprocessing
• HOG feature extraction
• Classification
In the preprocessing step, we have some basic image processing to separate characters from
real samples or preparing data from dataset and then in the second part, we extract HOG
features which is very distinguishable descriptor for character recognition where we divide an
input image into 99 cells and compute then the histogram of gradient orientations thereby
we represent each character with a vector of 81 features. The overall view of the proposed
approach has been illustrated in Figure 1. HOG is a fast and reliable descriptor which can
generate distinguishable features. Also, SVM is a fast and powerful classifier which is useful
to classify HOG features. The subsequent sections explain the steps in detail.
Figure 1. Steps in OCR.
801

3.1 Segmentation
The first step in processing is the segmentation of characters from the image. The image is
horizontally scanned to obtain the lines from image. Horizontal projection can be used to sepa-
rate lines since there will be ups and downs in the projection. Line segmentation is followed by
word and character segmentation. To segment lines into words and then to characters, vertical
projection can be used. Vertical projection will have gaps thereby separating the words and
individual characters.
3.2 Preprocessing
3.2.1 Noise removal
When a document is scanned, some noise is unavoidable. Noise may occur due to the quality
of the document, scanner etc. Before processing the image, the noise should be removed. A
low pass filter can be used to remove the noise thereby smoothing the image.
3.2.2 Binarization
Grayscale images are converted to two tone black and white images by the process called
binarization. In order to convert a grayscale image into a binary image, a value is selected as
the threshold and all pixel values above the threshold are converted to 1 and those below the
threshold are converted to 0.
3.3 HOG
HOG was first suggested (Dalal & Triggs, 2005) for detecting the presence of humans in
images. It also has wide applications in computer vision and image processing areas because
of its characteristics such as invariance to illumination and local geometric transformations.
The input image is partitioned into small square cells and then computes the histogram of
gradient directions or edge directions. The orientation of each pixel in a cell is quantized into
bins of histograms. Each bin represents an angle range between 0° and 360° or 0° and 180°.
A feature vector is formed by combining normalized histograms of each cell.
Figure 2. Sobel mask.
Figure 3. HOG of Malayalam character with varying cell size.
802

HOG is robust to illumination and local geometric changes as the local histograms have
been normalized based on the contrast. It is also independent of image size and captures
localized information. Compared to the well-known SURF and SIFT operators, a HOG
feature is computed across the entire image. Thus, for character recognition systems, HOG
features are often better than other structural features since it is invariant to local geometric
changes.
HOG features on some Malayalam characters have been illustrated in Figure 3 where a
Sobel operator is used to find out the horizontal and vertical component (Kamble & Hegadi,
2015). The Sobel mask is shown in Figure 2.
The gradient along the x and y axes can be estimated as follows:
G x ( u, v ) = H ∗ I ( u, v ) (1)
and
G y ( u, v ) = H T ∗ I ( u, v ) (2)
where I(u, v) is the cropped character. Gradient G(u, v) is calculated as:
G ( u,v ) = Gx2 ( u,v ) + G y2 ( u,v ) (3)
Orientation is measured as:
G y ( u,v )
θ ( u,v ) = tan −1 (4)
Gx ( u,v )
3.4 Classification
In the third stage, a linear multi-class support vector machine has been employed to clas-
sify characters. The optimization criteria of support vector machines are the marginal width
between two classes. If the width is higher, so the separability of patterns. The patterns that lie
on the soft margin are called support vectors and they characterize the classification function.
3.5 Post processing

Once the characters are recognized, semantic rules should be applied to correct errors caused
during classification, i.e. in the case of Malayalam, some characters such as dependent vowels
and signs should not appear at the beginning of a word. Also, dependent vowels will appear
only with consonants and independent vowels should not appear in-between a word.
4 CONCLUSION
This paper proposes a model for handwritten Malayalam character recognition using histo-
gram of oriented gradients as feature descriptor. HOG is invariant to local geometrical changes
and illumination. SVM is used as the classifier since it helps to achieve robust performance.
REFERENCES
Bhowmik, S., Roushan, M.G., Sarkar, R., Nasipuri, M., Polley, S. & Malakar, S. (2014). Handwrit-
ten Bangla word recognition using HOG descriptor. In Fourth International Conference of Emerging
Applications of Information Technology (EAIT), (pp. 193–197). New York, NY: IEEE.
803

Dalal, N. & Triggs, B. (2005). Histograms of oriented gradients for human detection. In IEEE Computer
Society Conference on Computer Vision and Pattern Recognition., 1, (pp. 886–893). New York, NY:
IEEE.
Ebrahimzadeh, R. & Jampour, M. (2014). Efficient handwritten digit recognition based on histogram of
oriented gradients and SVM. International Journal of Computer Applications, 104(9).
Elleuch, M., Hani, A., & Kherallah, M. (2017). Arabic handwritten script recognition system based on
HOG and Gabor features. International Arab Journal of Information Technology (IAJIT), 14.
Elleuch, M., Lahiani, H., & M. Kherallah (2015). Recognizing Arabic handwritten script using support
vector machine classifier. In 15th International Conference on Intelligent Systems Design and Applica-
tions (ISDA) (pp. 551–556). New York, NY: IEEE.
Iamsa-at, S. & Horata, P. (2013). Handwritten character recognition using histograms of oriented gradi-
ent features in deep learning of artificial neural network. In International Conference on IT Conver-
gence and Security (ICITCS), (pp. 1–5). New York, NY: IEEE.
Kamble, P.M. & Hegadi, R.S. (2015). Handwritten Marathi character recognition using R-HOG fea-
ture. Procedia Computer Science, 45, 266–274.
Kulkarni, R.L. (2017). Handwritten character recognition using HOG, COM by OpenCV and Python.
International Journal, 5(4).
Newell, A.J. & Griffin, L.D. (2011). Multiscale histogram of oriented gradient descriptors for robust
character recognition. In International Conference on Document Analysis and Recognition (ICDAR),
Qinyunlong, S. (2008). Handwritten character recognition using multi-resolution histograms of ori-
ented gradients. In 11th IEEE International Conference on Communication Technology. (pp. 715–717).
New York, NY: IEEE.
Ryu, J., Koo, H.I. & Cho, N.I. (2014). Language-independent text-line extraction algorithm for hand-
written documents. IEEE Signal Processing Letters, 21(9), 1115–1119.
Tikader, A. & Puhan, N. (2014). Histogram of oriented gradients for English-Bengali script recognition.
In International Conference for Convergence of Technology (I2CT) (pp. 1–5). New York, NY: IEEE.
804

A novel approach for the veracity and impact prediction of rumors
Anju Rose G. Punneliparambil & N.D. Bisna

ABSTRACT: Malignant or unplanned falsehood can be spread on social media and can
hazardously affect people and society. Models for automated verification of rumors are
already developed. But the impacts of rumors are neither analyzed nor predicted with these
rumor verification models. Impact prediction of rumor can be used for determining whether
the rumor to be responded to or not. This is very relevant for sudden situations in which a
rumor with large negative impact has to be addressed. As an extension to the veracity predic-
tion model, impact prediction is proposed.
1 INTRODUCTION
Social media services have enhanced the way individuals get data and news about current
occasions. Traditional news media just offers information to individuals. But online social
media such as Twitter and Facebook are platforms for people to impart information and
their insights about the news occasions. Users on Twitter create their profile and share their
status or opinions. These posts are known as tweets. Despite the fact that an extensive volume
of content is posted on Twitter, the majority of data is not valid or valuable in giving infor-
mation about the occasion. There can be noise, spam, ads and personal emotions in tweets.
This makes the quality of content on Twitter faulty.
1.1 What is a rumor?

Social psychology literature characterizes a rumor as a story or statement whose truth value
is unconfirmed or intentionally false (Allport & Postman, 1965). Rumors are spread on the
web. Fake stories and false claims affect people’s life in harmful ways. False bits of gossip are
harming as they cause public panic and social distress. For example, on 25 August 2015, a
rumor about “shootouts and kidnappings by drug gangs happening close to school in Ver-
acruz” spread through Twitter and Facebook. This caused chaos in the city including car col-
lisions, since individuals left their cars in the middle of road and hurried to get their children
from school. (Ma et al., 2016).
1.2 Motivation
Debunking rumors at an early stage of diffusion is especially significant to limiting their
unsafe impacts. To recognize bits of rumor from truthful occasions, people and organizations
have frequently depended on common sense and investigative journalism. Rumor revealing
sites like snopes.com and factcheck.org are such cooperative endeavors. Be that as it may, on
the grounds that manual confirmation steps are associated with such endeavors, these sites
are not complete in their topical scope and furthermore can have long debunking delay.
In the area of Natural Language Processing (NLP), there is some recent research analyz-
ing and deciding the truth value of social media content. There are several works are already
done to predict the veracity of rumor in Twitter. By predicting the impact of a false rumor
along with its veracity, the rumor can be stopped immediately before it makes further chaos.
805

1.3 Organization of paper
Section 2 presents some previous works on rumor prediction. Section 3 discusses the pro-
posed methodology. The method for impact analysis along with veracity prediction is han-
dled. Section 4 gives brief idea about the data set and the evaluation criteria. The paper
concludes in Section 5.
2 RELATED WORK
There are many works about predicting and analyzing the veracity of rumors. Some of the
major works are discussed here.
Ma et al. (2016) propose a technique for detecting rumors from microblogs with Recurrent
Neural Networks (RNN). It shows a strategy that learns continuous representations of microb-
log events for detecting rumors. The model depends on RNN for learning the hidden representa-
tions that catch the variety of contextual information of important posts after some time. Using
RNN, social context information of an event is demonstrated as a variable-length time series.
At the point when individuals are exposed to a rumor claim, they will forward the claim or com-
ment on it, in this way making a continuous stream of posts. In this method, both the temporal
and textual representations from rumor posts are learned under supervision. The model is also
effective for early detection of rumors, where satisfactory precision could be accomplished.
RumourEval is a SemEval shared task and is for identifying and handling rumors and
reactions to them, in text (Derczynski et al., 2017). Derczynski et al. propose a shared task
where members investigate rumors as cases made in content. In the same task, users react to
each other inside discussions attempting to determine the veracity of the rumor. If one needs
to evaluate the evidence of a rumor, different sources can be used to make a final decision
about the rumor’s veracity. SemEval consists of two sub tasks: (a) classifying rumor tweets
into support, deny or comment, and (b) veracity classification. Sub task A corresponds to
talk around cases to confirm or refute them using crowd response analysis. Sub task B relates
to the AI-hard problem of evaluating a claim. Overall, it is a tedious philosophy.
Real time rumor debunking is the proposal of Liu et al. (2015). It is an efficient procedure
for mining language features like individuals’ opinions, discovering witness accounts and
getting the underlying belief from messages. Use of sourcing, network propagation, cred-
ibility and other user and meta features help to expose rumors. Their contributions include:
(1) approach to automatically debunk rumors on social media; (2) an authentic rumor data-
base constructed on real data, and the process of its creation; and (3) an algorithm for pre-
dicting veracity in real time is potentially faster than human verification.
In Vosoughi et al. (2017) a rumor prediction approach named Rumor Gauge is discussed.
Identification of salient features of rumors on Twitter is done by looking at a few aspects
of diffusion: linguistics and the users involved. Comparison of each aspect with respect to
spreading of true and false rumors is made with Rumor Gauge.
A rumor signature can be formed by extracting the time series from these features. Then
using Hidden Markov Models (HMMs), the rumor signature can be classified as true or false.
This paper suggests an approach to predict the veracity of rumors in reasonable time and
with sufficient accuracy. So, an extension to this approach is possible to predict the impact of
false rumor. This is to resist the spreading of that rumor.
3 METHODOLOGY
This section discusses both veracity prediction and impact prediction. Veracity prediction is
the same as that in the method of Vosoughi et al. (2017).
3.1 Veracity prediction

A temporal communication network can be interpreted as a rumor in this case, where nodes
correspond to users, communication between nodes can be represented as edges in the
806

network. Time related features capture the propagation of messages through the network.
It notes that rumor features would be related to either the nodes or the edges in the net-
work. Features related to users are known as user identities and those related to messages as
linguistic.
3.1.1 Linguistic
Features from the text of the rumor are analyzed and those are collectively known as lin-
guistic features. Some linguistic features were identified. Features that significantly con-
tribute to the system are mentioned here. Significance can be found using a chi-square
test.
Ratio of negated tweets: This is the ratio of tweets having negation over the total number
of tweets in a rumor. Negations are detected using the Stanford NLP parser (Chen & Man-
ning, 2014).
Average maturity of tweets: Politeness and elegance considered as maturity of a tweet.
There are five indicators of the maturity of a tweet:
1. Smileys.
2. Abbreviations: Number of abbreviations (such as gn for goodnight, sry for sorry, gbu for
god bless you) present in the tweet.
3. Vulgarity: Number of vulgar words present in a tweet.
4. Word complexity: Length of words in the tweet are considered for checking the maturity
of the tweet.
5. Sentence complexity: Complexity of sentence contributes to the maturity of the tweet.
Ratio of tweets containing opinion and insight: Linguistic Inquiry and Word Count
(LIWC) gives a list of insight and opinion words. Words from the tweet are compared to
words in the category of LIWC dictionary (Pennebaker et al., 2003).
Ratio of uncertain and guessing tweets: Guessing and uncertain words come under another
category of LIWC. This includes words such as perhaps, like, guess, and so on. Each tweet is
checked against the guessing and uncertain words from LIWC.
3.1.2 User identities

User identity features consist of features of the user who is involved in spreading a rumor. A
total of six user features were identified which contribute to the output of model.
Controversiality: This is calculated by counting reactions to the user’s tweet. These reac-
tions are then run through a Twitter sentiment classifier (Vosoughi et al., 2015). It classifies
them as either positive, negative, or neutral. The count of all positive and negative reactions
is taken and based on that a controversiality score for the user can be calculated.
min( p / n ,n / p )
Controversiality = ( p + n ) (1)
In Equation 1, p denotes the number of positive reactions and n denotes the number of
negative reactions.
Originality: This is the ratio of the number of original tweets a user has posted to the
number of times the user posted retweets of someone else’s tweet.
Credibility: This checks that the user’s account has been officially verified by Twitter.
Influence: Influence is found by the number of followers of a user.
No of Followers
Role = (2)
No of Followees
T + Rt + Rp + F
Engagement = (3)
AccountAge
In Equation 4, T denotes the number of tweets, Rt denotes the number of retweets, Rp

denotes the number of replies and F denotes the number of favorites.
807

3.2 Impact prediction
Impact prediction is based on three characteristics. Response rate, profile feature and diffu-
sion rate are taken into account for predicting the impact of a false rumor.
3.2.1 Response rate

This can be further classified into positive and negative response rates. Positive Response
Rate (PPR) means what fraction of people respond to a false rumor in a way they think
that is true. Negative Response Rate (NPR) means the fraction of users who considered the
rumor as fake itself. PPR is proportional to the impact score, i.e. the higher the PPR, the
larger will be the impact of the rumor.
No of positive reactions
PPR = (4)
Total no of reactions
No of negative reactions
NPR = (5)
Total no of reactions
3.2.2 Profile analysis

This feature analyses the profile of the user and his/her followers. It is useful to check similar
kinds activities done in the past. If a previous action is found and it did not create any chaos,
then the rumor tweet may not be harmful. Like that history information (reactions of user to
similar rumors) is also helpful to predict the impact.
3.2.3 Diffusion rate

The diffusion rate is the number of reactions (i.e replies, likes, shares, etc.) per an hour. If the
diffusion rate is larger, the impact score will be larger too. That means a false rumor with a
high spreading tendency will have a very high negative impact on society.
3.3 Model
User identity features and linguistic features determine the signature of a rumor. Figure 1
depicts the overview of the proposed model. Some rumor tweets are manually annotated for
training of the model. Hidden Markov Model (HMM) is trained using annotated rumors. If
a new tweet arrives, the model compares the tweet with the stored collection. Then it predicts
the veracity. If the rumor veracity found to be less than 0.2, then it will be considered as fake.
Then the corresponding tweet is taken for impact analysis. A separate HMM is needed to
Figure 1. Overview of proposed methodology.
808

predict impact. False rumors from the database are used to train the HMM. Then an impact
score will be returned. The impact score is based on response rate, profile analysis and diffu-
sion rate. If the impact score is close to 1, then the rumor is considered to be harmful.
4 DATASET AND EVALUATION CRITERIA
4.1 Dataset
The dataset is available as a single JSON file. It includes 300,002 tweets about the health and
death of Tamilnadu chief minister, Jayalalitha. Every tweet that is associated with the situa-
tion is included in the dataset. For predicting the veracity of rumors, the model can also be
trained with PHEME dataset of rumors. It contains tweets about eight events. For example,
Ferguson unrest in US and the Germanwings plane crash in the French Alps. This dataset is
publicly available (Zubiaga et al., 2016).
4.2 Evaluation criteria

Wikipedia, Snopes.com, and FactCheck.org (sites that check reliability) are utilized to recover
the trusted confirmation sources. These facts considered for manual annotation. There were
three main criteria for the evaluation of the proposed model.
1. Precision value of the model in anticipating the rumor veracity before verification by
trusted authorities.
2. Contribution of features in predicting veracity of the rumor.
3. Accuracy of the model as a function of elapsed time since the start of the rumor.
5 CONCLUSION
Verification of rumors is a critical task and influential to populations so they can make deci-
sions based on the truth. This paper described a model for the verification of rumors and
prediction of the impacts. As a part of the impact analysis, impact features such as response
rate, profile feature and diffusion rate are identified. The ability to predict the impact of the
rumor along with its impact might be applied in the emergency services. News consumers and
journalists can use the proposed model in their field of work.
REFERENCES
Allport, G.W. & Postman, L.J. (1965). The psychology of rumor. New York, NY: Russell, Russell.
Chen, D. & Manning, C.D. (2014). A fast and accurate dependency parser using neural networks.
(pp. 740–750).
Derczynski, L., Bontcheva, K., M. Liakata, M., R. Procter, R., Hoi, G.W.S & Zubiaga, A. (2017).
Semeval-2017 task 8: Rumoureval: Determining rumour veracity and support for rumours. In SemE-
val@ACL.
Liu, Xiaomo, Nourbakhsh, Armineh, Li, Quanzhi, Fang, Rui, Shah, & Sameena (2015). Real-time
rumor debunking on twitter. Melbourne, Australia.
Ma, Jing, Gao, Wei, Mitra, Prasenjit, Kwon, Sejeong, Jansen, Jim, Wong, Kam-Fai, Cha, & Meeyoung.
(2016). Detecting rumors from microblogs with recurrent neural networks. In The 25th International
Joint Conference on Artificial Intelligence (IJCAI 2016).
Pennebaker, J.W., Mehl, M.R. & Niederhoffer, K.G. (2003). Psychological aspects of natural language
use: Our words, our selves. Annual Review of Psychology, 1, 547–577.
Vosoughi, S., Mohsenvand, M. & Roy, D. (2017). Rumor gauge: Predicting the veracity of rumors on
twitter. ACM Transactions on Knowledge Discovery from Data 4.
Vosoughi, S., Zhou, H. & Roy, D. (2015). Enhanced twitter sentiment classification using contextual
information. pp. 16–24.
Zubiaga, A., Liakata, M. & Tolmie, P. (2016). Pheme rumour scheme dataset: Journalism usecase.
809

A framework for efficient object classification for images having

noise and haze using deep learning technique
A. Bhavyalakshmi & M. Jayasree

Department of Computer Science and Engineering, Government Engineering College, Thrissur, Kerala, India
APJ Abdul Kalam Technological University, Kerala, India
ABSTRACT: Object category classification is one of the most difficult tasks in computer
vision because of the large variation in shape, size and other attributes within the same object
class. Also, we need to consider other challenges such as the presence of noise and haze,
occlusion, low illumination conditions, blur and cluttered backgrounds. Due to these facts,
object category classification has gained attention in recent years. Many researchers have
proposed various methods to address object category classification. The main issue lies in the
fact that we need to address the presence of noise and haze which degrades the classification
performance. This work proposes a framework for multiclass object classification for images
containing noise and haze using a deep learning technique. The proposed approach uses an
AlexNet Convolutional Neural Network structure, which requires no feature design stage for
classification since AlexNet extracts robust features automatically. We compare the perform-
ance of our system with object category classification without noise and haze using standard
datasets, Caltech 101 and Caltech 256.
1 INTRODUCTION
Object category classification is the task of classifying an object into its respective class
within a group of object classes. The most important step in this task is the choice of feature
extraction method. Obtaining key features from an image-containing object is a burdensome
task due to variation in attributes like shape, size and color within the same object class.
There exist many conservative feature extraction methods like Scale-Invariant Feature Trans-
form (SIFT) (Lowe, 1999) and Histogram Oriented Gradient (HOG) (Dalal & Triggs, 2005)
features for object recognition. To overcome the limitations of low-level features (Chan et al.,
2015), dictionary learning and deep learning were introduced in which features were learned
from the data itself instead of manually designing features.
In the above specified feature extraction techniques, most of the time is spent in deciding
ideal features for all classes of objects and in the selection of a suitable classification method.
Although these methods provide hopeful results, due to the inability of these methods to cap-
ture most compact and flawless features. Most advanced methods, like deep learning neural
networks, replace them with feature extraction for object classification. The main advantage
of using deep learning architecture is that it learns compact and flawless features automati-
cally. The main disadvantage of these methods is that they require a huge amount of data
and computation power (Hieu Minh et al., 2016).
At the end of 20th century and in first decade of the 21st century, neural networks have
shown satisfying results for the object classification problem (Cireşan et al., 2011; Jia et al.,
2009). But in recent years the old neural network structures have been replaced by new deep
and complex neural network architectures called deep learning neural networks (LeCun
et al., 2015). As specified earlier, the main disadvantage of these deep learning structures
is that they require extensive training with a huge amount of data to capture flawless and
compact features.
811

During recent years, different deep learning architectures have been developed. AlexNet
(Zeiler et al., 2014) and ZFNet (Zeiler et al., 2014) are two examples of such deep learning
networks. The AlexNet came top in the ImageNet challenge in 2012 but was outperformed
by other deep networks in the following years. Despite this defeat, AlexNet is still used for
many image classification problems.
Images are mostly corrupted by Gaussian noise during the acquisition of the image in the
camera. The presence of haze can also make the image unclear. So, object categorization in
the presence of noise and haze becomes an important issue. This paper deals with the clas-
sification of noisy and hazy images by a deep trained network model AlexNet.
This work is mainly divided into two experiments, object classification of noisy and hazy
images and object classification of images without noise and haze.
The rest of the paper is arranged as follows: Section 2 gives the literature survey;
Section 3 gives the overall system architecture of the proposed system; and finally,
Section 4 gives the conclusion.
2 RELATED WORK
Hieu Minh et al. (2016) suggested a technique for feature extraction which combines AlexNet
with a Recursive Neural Network (RNN) known as AlexNet-RNN. The AlexNet is used to
extract optimal features and which is then fed to the RNN. The RNN has a structure similar
to that of an ordinary neural network, the only difference is that it retains the weight learned
for a while.
Jian et al. (2016) used an average and weighted feature extraction method. First the behav-
ior of the features is investigated. In the next step, the most powerful features are chosen and
after that the average of these best features is used for classification. In weighted average
combination some of the features are omitted by assigning a value zero. Finally, all the fea-
tures are integrated into a k-NN framework for classification purpose.
Shengye et al. (2015) used a two-level feature extraction method for image classification. In
the first level, Bag of words (BOW) along with a spatial image pyramid is used to extract first
level features. During the second level, the features from the first level are extracted based on
dense sampling and spatial area sampling. The second technique adapts a multiple kernel
learning that can be used to fuse different image feature to obtain compact features. They
used the Caltech 256 and 15 scene datasets, obtaining an accuracy of 54.7% and 89.32% for
Caltech 256 and 15 scene respectively.
Transfer learning based on a deep learning approach was employed by Ling et al., (2015).
In recent years, deep neural networks have been used to extract the high level, most compact
and selective features from images that can be used for different tasks in computer vision.
As the image features are passed through different layers, the feature learning takes place at
different levels and feature learning in different levels represents different abstractions (Ling
et al., 2015). It is very difficult to apply deep learning techniques to vision problems as it
requires huge labeled data for training.
Foroughi et al. (2017) proposed an object categorization method which has significant
intra-class variation. It uses a joint projection and low-rank dictionary learning method
using dual graph constraints (JP-LRDL). It simultaneously learns a robust projection and a
discriminative dictionary in the low dimensional space. These can handle different types of
variation within the same object class, that raises due to occlusion, changes in viewpoint and
poses, size changes and various shape alterations.
Demir and Guzelis (2016) proposed a method for object recognition using two variations
of CNN. The first CNN included ten layers and the second one was similar to AlexNet
which consisted of nine layers. For feature extraction, the entire image is divided into nine
patches and features were extracted from each patch. The feature extraction method uses the
above specified CNN variants and BOW. The BOW uses SURF as the feature detector and
HOG as the feature descriptor. Finally, the features are supplied to an SVM for classifica-
tion. Existing methods do not consider the presence of various artifacts such as noise and
812

haze during classification. We are trying to implement a system that classifies objects even
in the presence of noise and haze using the deep convolutional neural network architecture,
AlexNet. And finally, compare the work with the object classification system with images
without any artifacts.
3 PROPOSED FRAMEWORK
In this section, we briefly summarize the essential steps for multiclass image category clas-
sification in the presence of noise and haze using a deep learning technique.
3.1 Object categorization

We conducted experiments using Caltech 101 and Caltech 256 images. Caltech 101 contains
101 different image classes (examples: faces, bags, watches, etc.). Each class has 40 to 800
images with 300 × 200 pixel size. Also, it contains background cluttered images. Caltech 256
is a successor of the Caltech 101 dataset. It overcomes certain limitations of Caltech 101 such
as the limited number of categories. It has a total of 30,607 images with 256 categories. The
minimum number of images per category is 80. This dataset is harder than the former one as
it has more categories and no left right aligned images. It also contains new and larger clutter
categories.
3.2 Training and testing sets

In the first step, the AlexNet framework extracted features from images in Caltech 101 and
Caltech 256 with no alterations. For this, we divided the full dataset into training and testing
sets. In the second phase different noises like Gaussian noise, salt and pepper noise and haze
were added to images. The images containing noises and haze were then randomly divided
into testing and training images. The training and testing set was used to extract training and
testing features from the AlexNet framework respectively. Finally, we compared the perform-
ance of these two methods.
3.3 Feature extraction using AlexNet

Convolutional Neural Networks (CNN) are a type of artificial neural networks that are
mainly used for object detection (Wanli et al., 2016), object recognition (Ming & Xiaolin,
2015) and image retrieval (Hailiang et al., 2017). Recently many researchers have used CNN
for solving other problems. The main layers in CNN include the convolution layer, the pool-
ing layer, the normalization layer and the fully connected layer. The main advantage of CNN
is that it extracts robust features automatically and we can reduce the dimension of features
learned by increasing the size of the hidden layers. Due to the weight sharing mechanism, we
require the minimum amount of parameters and neurons for training the CNN.
AlexNet is a variant of CNN which provides high performance and learns compact and
robust features. The basic AlexNet framework proposed by Alex et al. (2012) consists of five
convolution layers and three max pooling layers. As in any CNN, the learning takes place in
Figure 1. Structure of AlexNet framework.
813

Figure 2. Architecture of the proposed framework.
the convolution layer. The pooling in a CNN can be either max pooling or average pooling.
For most of the image classification problems, max pooling is found to be more effective than
average pooling. Hence in this paper, we adopt max pooling.
The AlexNet used in this paper consists of one input layer, five convolution layers, three
pooling layers, three fully connected layers, along with two normalization layers, seven Rec-
tified Linear Unit (ReLU) layers, two dropout layers and a softmax and output layer. The
intermediate layers: convolution, pooling, fully connected and ReLU layers form the bulk of
the CNN. The convolution layer performs the convolution operation which is the same as con-
volution in image processing. What it actually does is remembers the pattern learned during
the training process and when a new pattern occurs it tries to label it to the closest pattern. The
pooling layer is used to reduce the feature size. It down samples the feature map by dividing it
into a rectangular pooling region and selecting the maximum value from each region. When
using CNN for image classification, the choice of activation function is very important. Usu-
ally, we can use the hyperbolic tangent function (tanh) or ReLu. In our model, we use ReLu
as the activation function. In this framework, we use a max pooling layer with size 3 × 3 and
stride 2. The first convolution layer uses 96 kernels with size 11 × 11@3, the second uses 256
kernels with size 5 × 5@48, the third and fourth uses 384 kernels with size 3 × 3@256 and the
final one uses 256 kernels with size 3 × 3@192. The first two convolution layers are followed
by a normalization, ReLu and max pooling layer. The next two are only followed by a ReLu
layer. The final convolution layer is followed by one ReLu and pooling layer. After the sixth
and seventh ReLu layers we employed two dropout layers of 50% to avoid overfitting.
3.4 Classification
We can use different classifiers such as k-Nearest Neighbor (k-NN) (Yin & Bo, 2009), Naive
Bayes (Shih-Chung et al., 2015), Hidden Markov Model (HMM) (Jia et al., 2000) and Sup-
port Vector Machine (SVM) (Cortes & Vapnik, 1995).
814

In our framework, we use SVM for classification. SVM is a supervised machine learning
algorithm which is generally used for both classification and regression challenges. SVM uses
a hyperplane or set of hyperplanes for classification and regression. Since it tries to maximize
the margin size, it is also called maximum margin classifier. The main advantage of using
SVM for classification is that it performs classification effectively even if the dimension is
high. The responses from the AlexNet are reshaped into feature vectors and passed to the
classifier for training and testing purposes. The main reason for using SVM for classification
is due to the overfitting issue in the fully connected layers and SVM also only requires a few
parameters compared to the fully connected layer for classification purposes.
4 CONCLUSION
In this paper, we have proposed an efficient method to classify a large number of object
categories in the presence of noise and haze. Here we use a variant of CNN, AlexNet, for
extracting flawless and compact features.
This approach requires no traditional feature extraction during the training stage as CNN
learns the features automatically, and classification can be performed very efficiently even in
the presence of noise and haze. The output of AlexNet is compact and highly relevant fea-
tures. Finally, we can use simple multiclass classifiers like SVM for effective classification.
Training and testing was conducted on two benchmark datasets Caltech 256 and Caltech
101, with noise and haze. We assume that there will be no considerable changes in the per-
formance even if artifacts like noise and haze are present.
REFERENCES
Alex, K., Ilya, S. & Geoffrey, E.H. (2012). ImageNet classification with deep convolutional neural net-
works. In Pereira, F., Burges, C., Bottou, L., Weinberger, K., (Eds.), Advances in Neural Information
Processing Systems 25 (pp. 1097–1105). Curran Associates, Inc.
Cireşan, D.C., Meier, U., Masci, J., Gambardella, L.M. & Schmidhuber, J. (2011). Flexible, high per-
formance convolutional neural networks for image classification. pp. 1237–1242.
Cortes, C. & V. Vapnik (1995, Sep). Support-vector networks. Machine Learning 20(3), 273–297.
Dalal, N. & Triggs, B. (2005). Histograms of oriented gradients for human detection. IEEE Computer
Society Conference on Computer Vision and Pattern Recognition 1, 886893.
Demir, Y. & Guzelis, C (2016). Moving towards in object recognition with deep learning for autono-
mous driving applications. IEEE Conference Publications.
Foroughi, H., Ray, N. & Zhang, H. (2017). Object classification with joint projection and low-rank
dictionary learning. IEEE Transactions on Image Processing 99, 1–1.
Hailiang, L., Yongqian, H. & Zhijun, Z. (2017). An improved faster r-cnn for same object retrieval.
IEEE Access 5, 13665–13676.
Hieu Minh, B., L. Margaret, L., N. Eva, Cheng. Katrina, & B. Ian S. (2016). Object recognition using deep
convolutional features transformed by a recursive network structure. IEEE Access 4, 10059–10066.
Jia, D., Wei, D., Richard, S., Li-Jia, L., Kai, L. & Li, F.-F. (2009). Image Net: A large-scale hierarchical
image database. IEEE Conference on Computer Vision and Pattern Recognition, 248–255.
Jia, L., N. A, & G.R. M (2000). Image classification by a two-dimensional hidden Markov model. IEEE
Transactions on Signal Processing, 48, 517–533.
Jian, H., Huijun, G., Qi, X. & Naiming, Q. (2016). Feature combination and the knn framework in
object classification. IEEE Transactions on Neural Networks and Learning Systems, 27, 1368–1378.
LeCun, Y., Bengio, Y. & Hinton, G. (2015). Deep learning. Nature, 521, 436444.
Ling, S., Fan, Z. & Xuelong, L. (2015). Transfer learning for visual categorization: A survey. IEEE
Transactions on Neural Networks and Learning Systems, 26, 1019–1034.
Lowe, D.G. (1999). Object recognition from local scale-invariant features. The Proceedings of the Sev-
enth IEEE International Conference, 2, 11501157.
Ming, L. & Xiaolin, H. (2015). Recurrent convolutional neural network for object recognition. IEEE
Conference on Computer Vision and Pattern Recognition (CVPR), 3367–3375.
815

Shengye, Y., Xinxing, X., Dong, X., Stephen, L. & Xuelong, L. (2015). Image classification with densely
sampled image windows and generalized adaptive multiple kernel learning. IEEE Transactions on
Cybernetics, 45, 381–390.
Shih-Chung, H., I-Chieh, C. & Chung-Lin, H. (2015). Image classification using pairwise local observa-
tions based naive Bayes classifier. Asia-Pacific Signal and Information Processing Association Annual
Summit and Conference (APSIPA), 444–452.
T.H. Chan, K. Jia, S.G.J.L.Z.Z. & Y. Ma. (2015). Pcanet: A simple deep learning baseline for image
classification. IEEE Transactions on Image Processing, 24, 50175032.
Wanli, O., Xingyu, Z., Xiaogang, W., Shi, Q., Ping, L., Yonglong, T., … & T. (2016). DeepID-Net:
Object detection with deformable part based convolutional neural networks. IEEE Transactions on
Pattern Analysis and Machine Intelligence, 39, 1320–1334.
Yin, L. & Bo, C. (2009). An improved k-nearest neighbor algorithm and its application to high resolu-
tion remote sensing image classification. 17th International Conference on Geoinformatics, 1–4.
Zeiler, M.D., e. D. Fergus, R., Pajdla, T. Schiele, B. & Tuytelaars, T. (2014). Visualizing and understand-
ing convolutional networks, book Title = Computer Vision – ECCV 2014: 13th European Confer-
ence, Zurich, Switzerland, September 6–12, 2014, Proceedings, Part I. Cham: Springer International
Publishing.
816

Bilingual handwritten numeral recognition using convolutional

neural network
Jettin Joy & M. Jayasree

ABSTRACT: Optical character recognition (OCR) is the process of identifying characters

in any form, with the help of a photo-electric device and computer software. Handwritten
numeral recognition can be considered as a special area in OCR, where the characters are
handwritten numerals. Unlike printed numeral recognition, handwritten numeral recogni-
tion is very difficult as different persons have different writing styles. Usually, Conventional
OCR includes steps like pre-processing, segmentation, feature extraction and finally clas-
sification and recognition. Conventional OCR uses handcrafted features for feature extrac-
tion, whereas Convolutional Neural Networks (CNN) extracts features automatically. Hence
by using CNN, we can eliminate the need for manual feature engineering. Recently many
researchers have used deep learning architectures like CNN, Stacked or Deep Autoencod-
ers and Deep Belief Networks for Handwritten Numeral Recognition and they have shown
promising results compared to other methods. No one has yet implemented a system for
recognizing Malayalam and Kannada handwritten numerals using CNN. So this is the first
project to address Malayalam and Kannada handwritten numeral recognition using deep
learning approach. Also, we compare the performance with another deep learning archi-
tecture, stacked autoencoder. The sole aim of this project is to automatically recognize and
classify Malayalam and Kannada numerals using Convolutional Neural Network.
1 INTRODUCTION
Optical character recognition (OCR) is one of the old and most popular research area.
Researchers are trying to improve the performance of existing methods by introducing new
methods in segmentation, feature extraction and classification. Handwritten character and/
or numeral recognition is a part of OCR where the characters are handwritten in nature.
Unlike printed numeral recognition, recognition of handwritten numerals is very difficult.
This is mainly due to the variation in the appearance of numerals, as different persons have
different writing styles.
Furthermore, handwritten character and/or numeral recognition can be either online or
offline. The online method involves recognition of characters in real-time i.e, the characters
are recognized as soon as they are written, usually the input is coordinate values of the char-
acters. On the other hand offline character recognition system takes input in the form of the
image, which is obtained from scanning, pre-processing and segmenting the document con-
taining the characters. Offline method is more challenging than the online as it involves noise
and other artifacts like document quality, variation in the shape and style of characters etc.
Recently deep learning methods have shown promising performance in areas like object
recognition (Ayegl et al. 2016), object classification (Wanli et al. 2016), automatic number
plate recognition (Menotti, Chiachia, Falco, & Oliveira Neto 2014, Syed Zain, Guang,
Afshin, & Enrique 2017), sentiment analysis in texts (Abdalraouf & Ausif 2017), stock
market prediction (Vargas et al. 2017), automatic speech recognition (Palaz et al. 2015) and
character recognition (Mehrotra et al. 2013). From these we can conclude that nowadays
817

Figure 1. Malayalam and Kannada numerals.
deep learning methods are used in almost all research areas such as Data analytics, natural
language processing (NLP), Object and Image classification and character recognition etc.
Convolutional Neural Network or CNN or ConvNet is a type of deep learning neural net-
work architecture, that is most suitable for image structure representation and visual imagery
analysis. The main advantage of CNN is that it requires minimal amount of pre-processing
(Lecun et al. 1998).
In recent years several variants of Convnet has emerged in the ImageNet challenge, some of
them include Alex Net (Alex et al. 2012), ZF Net (Zeiler, Fergus, Pajdla, Schiele, & Tuytelaars
2014), Google Net (Christian, Wei, Yangqing, Pierre, Scott, Dragomir, Dumitru, Vincent, &
Andrew 2015) and Res Net (He et al. 2015). The main advantage of using deep learning
structure like ConvNet is that we can eliminate the need for manual feature extraction. Deep
neural networks can outperform the conventional methods (P Nair, Ajay, & Chandran 2017).
Malayalam and Kannada are two of the most commonly used languages in South India
and are the official languages of the Kerala and Karnataka state respectively. These two lan-
guages belong to Dravidian language family. Malayalam and Kannada language have their
own numerals and are more complex compared to modern Hindu-Arabic numerals due to
the complex curved nature. Figure 1 shows Malayalam and Kannada Numerals.
To my extent of knowledge, this is the first paper to address Malayalam and Kannada
handwritten numeral recognition using Convolutional Neural Network. We are attempting
to use CNN to achieve better performance in recognizing Malayalam and Kannada hand-
written numerals.
2 LITERATURE SURVEY
Akm and Abdul (2017) used CNN for Handwritten Arabic numeral recognition. They con-
ducted experiment with two models. In the first model, they used they used a variant of multi
layer perceptrons (MLP) with notable changes. To avoid over-fitting, they employed dropout
mechanism. The second is a ConvNet model with two convolution and two max pooling
layer. The first convolution layer includes 30 feature maps each with kernel size 55. The sec-
ond convolution layer include 15 feature maps each with kernel size 3 × 3 pixel. Both the max
pooling layer had a size of 22. In order to avoid over-fitting a dropout of 25% was used. The
accuracy for the first model was 93.8% and that of second model (CNN Model) was 97.4%.
Akhand et al. (2015) used a CNN based architecture for recognizing Bangla handwritten
numerals. The CNN included 2 convolution layer with 5 × 5 kernel and 2 subsampling layer with
size of 2 × 2, which uses average pooling. The first level feature feature maps (6 feature maps)
that obtained after first convolution had a size of 24 × 24. Using the first subsampling layer the
feature maps were downsampled to 12 × 12 and the second convolution produced 12 feature
maps of 8 × 8. Finally, using a second level subsampling it is further downsampled to 4 × 4. No
dropout mechanism were used in this method. The overall accuracy for the system was 97.93%.
Ramadhan et al. (2016) used a similar CNN structure as above for handwritten mathemat-
ical symbol recognition, they used 3 feature map in the first convolution and subsampling
and 5 feature maps in the second convolution and subsampling and unlike average subsam-
pling used above they employed max pooling. The overall accuracy for training and testing
was 93.27% and 87.72%. respectively.
818

Another deep learning architecture like deep sparse belief network and stacked denoising
autoencoders are also used for handwritten digit recognition (Ragheb and Ali 2014). The spare
deep belief network consists of three RBMs with 600 hidden units. The training was done in
a greedy manner using contrastive divergent method. They also proposed a stacked denoising
autoencoder architecture with a corruption rate of 25% and 600 units each hidden layer.
Md et al. (2016) proposed a structure with an autoencoder with 3 convolution layer and
3 max pooling layers, a decoder with 3 deconvolutional layer and 3 upsampling layers. The
output of this is passed through a Convolutional layer and finally the output is obtained.
They used 3 models for training. The first one is called SCM: Simple convolutional model,
the second one is called SCMA: simple convolutional model with augmented images. The
final one is called ACMA: Autoencoder with Convolutional Model with Augmented images.
And they found that the third model has high efficiency than the former two.
Pal et al. (2007) proposed a system for recognizing handwritten numerals of six Indian
scripts. They used two type of feature vectors, first one is of length 64 and is used for fast rec-
ognition. The second one is of length 400 and is used for higher accuracy recognition. They
used contour points to extract the first features and gray-scale local-orientation histogram
of the component to extract second features. The system showed satisfactory performance
for all languages.
V Rajashekararadhya and Vanaja Ranjan (2009) proposed a zoning and distance based fea-
ture for effectively recognizing Kannada numerals with support vector machine (SVM) clas-
sifier. They divided entire image into n zones after calculating the centroid. After that average
distance from each pixel to the centroid is calculated and repeated for each zone. and a total of
n features are extracted for classification and recognition. Finally, a Support vector machine
is used as classifier. The overall performance listed was satisfactory for all class of numerals.
The above two methods use handcrafted feature extraction method.
We are attempting to implement a efficient system for recognizing Malayalam and
Kannada numerals using CNN that uses automatic feature extraction methods to improve
the existing performance.
3 PROPOSED SYSTEM
The Convolutional Neural Networks need to trained with large image dataset to obtain good
performance. This is the main challenge in any deep learning project. i.e, it requires a large
dataset. However, there are techniques to increase the size of images in the dataset. There is
no open source large dataset available for Malayalam and Kannada numerals. Figure 2 shows
the overall architecture of the proposed system.
The proposed system includes the following steps:
3.1 Data acquisition and dataset creation

There is no open source large dataset for numeral recognition in Malayalam and Kannada.
Hence we require to build it from the scratch. This is the most time consuming and tedious task.
In this step, we collect the data from a large population. As CNN requires large dataset, it is
important to collect as many numeral writings from different persons. After collecting data the
document containing numeral images are scanned and individual numerals are segmented from
the scanned image using appropriate segmentation mechanism. The Figures 3 and 4 shows
three sets of samples of Malayalam and Kannada handwritten numerals respectively.
3.2 Pre-processing
The pre-processing step usually involves resizing, normalization and removing unwanted enti-
ties from numeral images, like noise, haze etc. Pre-processing steps helps to ease the numeral
recognition. Firstly, the images are resized to an appropriate size, then it is converted to gray
scale image. After that, the pixels in images are inverted to obtain the negative of the image.
819

Figure 2. System architecture.
Figure 3. Malayalam handwritten numerals.
Figure 4. Kannada handwritten numerals.
This is usually done to reduce the storage size. The image size has great influence in the train-
ing stage. If it is too large we will require increased training time as computation involved is
high. If the size is too small fitting of images into the network becomes difficult, hence it is
820

Figure 5. LeNet-5 architecture.
always good to choose an appropriate size. We can use methods like padding for choosing a
standard size.
3.3 Dataset augmentation

As mentioned earlier, CNN requires large image dataset for training. To obtain large dataset
we use dataset augmentation. For augmenting the images affine transformations like trans-
lation, scaling, rotation and shearing can be used. The affine transforms mentioned above
generally preserves points, straight lines, and planes. Other methods like blurring, vary the
contrast and brightness of the images can be applied to obtain a large dataset from the exist-
ing dataset. After the augmentation step, the entire dataset is randomly divided into testing
images and training images, e.g 70% for training and remaining 30% for testing.
3.4 CNN modeling

This is the most vital step in this project. Here we define various layers in CNN and appropriate
network parameters for training the dataset. The main layers in a CNN are described below.
3.4.1 Convolution layer

This is the first layer that receives input, the convolution operation tries to label a new image
class by referring what it has learned in the past. The main property of convolution operation
is that it is translation invariant (Thomas & Helmut 2015).
3.4.2 Subsampling layer

The subsampling layer takes input from the convolution layer. It is used to smooth or reduce
the feature map size or feature size. pooling can be done in two ways. First one is called max
pooling and the later one is called average pooling, for most of the problems max pooling is
found to be more effective than average pooling, Hence we adopt max pooling mechanism.
Figure 5 shows a simple LeNet (Lecun et al. 1998) CNN architecture.
3.4.3 ReLu layer

This layer mainly determines which activations are propagated to next layer. Usually, output
signals that are more close to previous references are propagated. There are many activation
functions that can be used in which ReLu is most effective for fastening the training process.
3.4.4 Fully connected layer

This is the final layer in CNN, which means that every neuron of the current layer receives out-
put from the previous layer. The layers involved in feature learning are Convolution layer, Sub-
sampling layer, and ReLu layer. After this phase, a CNN model with different layer is obtained.
3.5 Training and testing

The CNN model is used for training and testing the dataset. In order to avoid overfitting,
dropout mechanism can be used i.e, dropping out a certain number of neuron output in each
layer (Srivastava, Hinton, Krizhevsky, Sutskever, & Salakhutdinov 2014). The testing images
821

are used to test the model created. A randomization technique is used for dividing the entire
dataset into training and testing set.
3.6 Classification
This is the final stage, in which a softmax function is used to classify the output of the CNN.
The softmax function output falls within the range [0,1] and the sum of the output of all class
is equal to 1. The softmax function classifies the input numeral to a class that has highest
output value. The above system can be evaluated by varying the layers in CNN and also vary-
ing different parameters like learning rate. Backpropagation with gradient descent is most
effective learning rule for image classification problem.
4 CONCLUSION
Handwritten numeral recognition has a large number of applications like ZIP code recogni-
tion (LeCun et al. 1989), recognizing numerals in old documents etc. Traditional methods use
handcrafted features, which requires a great deal of effort and time. This can be eliminated
by introducing new automatic feature learning methods like Convolutional Neural Network,
Deep Belief Network, AutoEncoders etc. These deep learning method has shown outstand-
ing performance in recognizing numerals as well as handwritten characters (Shailesh et al.
2015).
Here we have proposed a bilingual handwritten numeral recognition system for Malayalam
and Kannada numerals. The dataset creation and CNN modeling is very time consuming.
We would provide first ever large open source dataset for Malayalam and Kannada numerals.
To reduce the time required for the training stage graphics processing unit (GPU) support is
used. Also to avoid overfitting in training phase of the CNN dropout mechanism is applied.
CNN have shown better results in recognizing numerals in various scripts, hence has a high
probability that it will provide the same for Malayalam and Kannada numerals.
REFERENCES
Abdalraouf, H. & M. Ausif (2017). Deep learning approach for sentiment analysis of short texts. 3rd
International Conference on Control, Automation and Robotics (ICCAR), 705–710.
Akhand, M.A.H., M. Mahbubar Rahman, P.C. Shill, I. Shahidul, & M.M. Hafizur Rahman (2015).
Bangla handwritten numeral recognition using convolutional neural network. International Confer-
ence on Electrical Engineering and Information & Communication Technology (ICEEiCT2015), 1–5.
Akm, A. & K.T. Abdul (2017). Handwritten arabic numeral recognition using deep learning neural
networks. CoRR abs/1702.04663.
Alex, K., S. Ilya, & E.H. Geoffrey (2012). Imagenet classification with deep convolutional neural net-
works. In Pereira, F., Burges, C., Bottou, L., Weinberger, K., eds.: Advances in Neural Information
Processing Systems 25. Curran Associates, Inc., 10971105.
Ayegl, U., D. Yakup, & G. Cneyt (2016). Moving towards in object recognition with deep learning for
autonomous driving applications. International Symposium on Innovations in Intelligent SysTems and
Applications (INISTA), 1–5.
Christian, S., L. Wei, J. Yangqing, S. Pierre, R. Scott, A. Dragomir, E. Dumitru, V. Vincent, &
R. Andrew (2015). Going deeper with convolutions. IEEE Conference on Computer Vision and Pat-
tern Recognition (CVPR).
He, K., X. Zhang, S. Ren, & J. Sun (2015). Deep residual learning for image recognition. CoRR 7.
LeCun, Y., B. Boser, J.S. Denker, D. Henderson, R.E. Howard, W. Hubbard, & L.D. Jackel (1989).
Backpropagation applied to handwritten zip code recognition. Neural Comput. 1(4), 541–551.
Lecun, Y., L. Bottou, Y. Bengio, & P. Haffner (1998). Gradient-based learning applied to document
recognition. 86, 2278–2324.
Md, S., M. Nabeel, & A. Md Anowarul (2016). Bangla handwritten digit recognition using autoen-
coder and deep convolutional neural network. International Workshop on Computational Intelligence
(IWCI), 64–68.
822

Mehrotra, K., S. Jetley, A. Deshmukh, & S. Belhe (2013). Unconstrained handwritten Devanagari charac-
ter recognition using convolutional neural networks.
Menotti, D., G. Chiachia, A.X. Falco, & V.J. Oliveira Neto (2014). Vehicle license plate recognition with
random convolutional networks. 27th SIBGRAPI Conference on Graphics, Patterns and Images 249,
1530–1834.
Nair P., P., J. Ajay, & S. Chandran (2017). Malayalam handwritten character recognition using convo-
lutional neural networks. International Conference on Inventive Communication and Computational
Technologies (ICICCT), 278–281.
Pal, U., N. Sharma, T. Wakabayashi, & F. Kimura (2007). Handwritten numeral recognition of six
popular indian scripts. Ninth International Conference on Document Analysis and Recognition, 2007,
ICDAR 2007 2, 749–753.
Palaz, D., M. Magimai-Doss, & R. Collobert (2015). Convolutional neural networks-based continuous
speech recognition using raw speech signal. IEEE International Conference on Acoustics, Speech and
Signal Processing (ICASSP), 4295–4299.
Ragheb, W. & L. Ali (2014). Handwritten digit recognition using sparse deep architectures. 9th Interna-
tional Conference on Intelligent Systems: Theories and Applications (SITA-14), 1–6.
Rajashekararadhya, V., S. & P. Vanaja Ranjan (2009). Support vector machine based handwritten
numeral recognition of kannada script. IEEE International Advance Computing Conference, 2009.
IACC 2009., 381–386.
Ramadhan, I., B. Purnama, & S. Al Faraby (2016). Convolutional neural networks applied to handwrit-
ten mathematical symbols classification. 4th International Conference on Information and Communi-
cation Technology (ICoICT), 1–4.
Shailesh, A., P. Ashok Kumar, & G. Prashnna Kumar (Software, Knowledge, Information Management
and Applications (SKIMA), 2015). Deep learning based large scale handwritten devanagari charac-
ter recognition. 9th International Conference on Software, Knowledge, Information Management and
Applications (SKIMA), 2015, 1–6.
Srivastava, N., G. Hinton, A. Krizhevsky, I. Sutskever, & R. Salakhutdinov (2014). Dropout: A simple
way to prevent neural networks from overfitting. J. Mach. Learn. Res. 15, 1929–1958.
Syed Zain, M., S. Guang, D. Afshin, & G.O. Enrique (2017). License plate detection and recognition
using deeply learned convolutional neural networks. CoRR abs/1703.07330, 1703.07330.
Thomas, W. & B. Helmut (2015). A mathematical theory of deep convolutional neural networks for
feature extraction. CoRR abs/1512.06293.
Vargas, M., B. Lima, & A.G. Evsukoff (2017). Deep learning for stock market prediction from financial
news articles. IEEE International Conference on Computational Intelligence and Virtual Environments
for Measurement Systems and Applications (CIVEMSA), 60–65.
Wanli, O., Z. Xingyu, W. Xiaogang, Q., Shi, L. Ping, T. Yonglong, L. Hongsheng, Y. Shuo, W. Zhe,
L. Hongyang, Kun, W., Y. Junjie, L. Chen-Change, & T. (2016). Deepid-net: Object detection with
deformable part based convolutional neural networks. IEEE Transactions on Pattern Analysis and
Machine Intelligence 39, 1320–1334.
Zeiler, M.D., e. D. Fergus, Rob., T. Pajdla, B. Schiele, & T. Tuytelaars (2014). Visualizing and Under-
standing Convolutional Networks, book Title=Computer Vision – ECCV 2014: 13th European Confer-
ence, Zurich, Switzerland, September 6–12, 2014, Proceedings, Part I. Cham: Springer International
Publishing.
823

A security framework to safeguard the hybrid model of SDN from

malicious applications
C.M. Mansoor & K.V. Manoj Kumar

ABSTRACT: The hybrid control model of SDN is designed to improve network produc-
tivity by reducing the controller load. When the network is under heavy load, flow rules are
installed on a network device by other network devices on behalf of the controller, and in the
case of normal load the control is centralized. Thus the controller does not have to program
flows to each network equipment one by one, instead it can ask the equipment to spread this
flow to other equipment on behalf of the controller. This model is not secure from malicious
applications as all the applications are treated in the same way and there is no way to distin-
guish between a genuine and a malicious application. This paper proposes a permission sys-
tem to which the applications must subscribe on initialization with the controller and before
approving the application commands a permissions check is performed. The priority of the
application is also considered while granting the permission in order to deal with policy con-
flict.This will effectively monitor the working of every application and thus will prevent any
unauthorized operations.
1 INTRODUCTION
The main feature of the SDN architecture is the separation of the control plane from the data
plane. The data plane devices forward packets on the basis of instructions obtained from a
logically centralized controller which also maintains the network state. In traditional networks
a change in the device configuration or the routing strategy would mean the modification of
the firmware of all the involved data plane devices, which would incur high cost. Since SDN
implements the control plane in software, the changes in the routing strategy can be made
from a single point i.e. there is centralization of policies. The forwarding devices no longer
needs to make decisions and are thus less complex allowing the creation of low cost devices.
Open flow is the most widely used protocol for communication between the control and
data plane. In an attempt to efficiently balance the controller load when using open flow, Oth-
man & Okamura (2013a) proposed a hybrid control model of SDN; that allows the regular
centralized control model to be used in the normal situations, but at the same time introduces
a distributed control model in order to ensure proper working of the network in situations
where the controller is under substantial loads and is required to install large number of flows
in to the forwarding devices. In such cases, the hybrid control model can be used to relieve the
controller from doing any further processing to relocate the flows, and enabling it to install
those flows as they are; and relying on the distributed control of the network equipment to
solve any issues of network equipment overloading. And thus, the hybrid control model ena-
bles the smooth working of the network even in cases of overloading.
The SDN architecture provides both advantages and disadvantages to the security of the
network. On one hand, once an attack is detected the global view of the network allows to
take countermeasures much quickly and on the other hand it brings with it additional security
challenges. One of the issues is the attack by malicious applications. Vulnerability created in
the network by granting complete control and visibility of the network to the applications was
discussed in X. Wen & Wang. (2013). The interaction of the applications with the network
825

can be monitored by the use of a permissions set which works in the same way as the system
used in android play store as shown in S. Scott-Hayward & Sezer (2014). In the Android smart
phone scenario, various applications will only be granted minimal set of permissions required
for its working. In the same way SDN application designed to read the network topology
should not be granted permission to obtain various real time notifications which would make
the network vulnerable by exposing sensitive information to the controller.
This paper proposes a security application on top of the hybrid control model of SDN,
to safeguard the network from malicious applications by using a permission system.Applica-
tions can subscribe to various permissions and can only execute only those operations for
which permission has been granted. This allows to prevent the malicious operations and to
monitor the working of every application.
2 PROBLEM DESCRIPTION
In the hybrid model of SDN, the applications interact with the network via the controller
which provides an abstraction of the data plane elements. The applications can get informa-
tion such as the current state of the network or can change the routing strategy by asking the
hybrid controller to write the flows to various devices. The controller will either spread the
flow by itself or in case of heavy load ask other equipment to spread the flows. Every applica-
tion is treated in the same way and there is no distinction between a genuine application and
a malicious application.
Although the effect of some attacks such as dos will be less severe in this model compared
to the normal SDN, an attacker can control the entire network by writing the intended flow
rules to the network devices via a malicious application. Also there is no control over the type
of operations that genuine applications can perform.
In addition to securing the hybrid model of SDN by preventing malicious applications
from taking control over the network, this work also enforces priority to different applica-
tions to ensure that their operations do not interfere with each other.
3 RELATED WORK
Various security issues and corresponding solutions were discussed in Sandra Scott-Hayward &
Sakir Sezer (2016) The problem of treating every application with the same privilege was first
identified in X. Wen & Wang. (2013). The authors propose PermOF with a set of permissions
which were enforced at the Application Programming Interface (API) entry using an isolation
mechanism. It was successful in solving the app privilege problem and thus securing the net-
work. The concept of the permissions system is extended in S. Scott-Hayward & Sezer (2014). It
is implemented on top of the Floodlight controller. The authors define the set of permissions to
which the application must subscribe on initialization with the controller and introduce an Oper-
ation Check point, in which before approving the application commands a permissions check
is performed. They also used an unauthorized operations log to examine the malicious activity
in order to build a profile for SDN application—layer attacks. Although it discussed about the
problem of application priority enforcement, a solution for that specific issue was not presented.
The issue of policy conflict was discussed in P. Porras & G. Gu (2012). The system uses
the FortNOX enforcement engine, which handles possible conflicts by using authors security
authorization to decide on flow insertion. It checks whether the new flow clashes with the
existing flow rule. If the new flow rule is conflicting with the existing one, it will be installed
only if it is issued by a higher priority author. The need to resolve the proper authorization
level of the flow rule author is a drawback of this method.
Stanford research institute extended the floodlight Controller to develop the Security
enhanced floodlight (SEK). An administrator authorizes applications java class, which is
digitally verified by the SEK at run time. The application has full control over the network
once it is signed and approved.
826

In an attempt to improve network productivity by reducing the controller load, a hybrid
control model was discussed in Othman & Okamura (2013a). In order to ease the control-
ler pressure when the network is under heavy load, flow rules are installed on a network
device by another network device on behalf of the controller, and in the case of normal
load, the control is centralized. The same authors presented the method to secure the
distributed control model in Othman & Okamura (2013b). The central control element is
secured by TLS and the transmission of flow installation requests between the network
devices is protected using a signature algorithm. The system also uses a centralized trust
manager. The solution in the present paper will be implemented on top of Othman &
Okamura (2013b).
4 DESIGN
A complete set of permissions is defined to which the applications must subscribe on ini-
tialization with the controller.The permission set is similar to that in X. Wen & Wang. (2013)
and it includes all types of permissions an application may need to execute. The permissions
are stored securely with the application IDs linked to the group of permissions allowed to
an application similar to S. Scott-Hayward & Sezer (2014) and the network administra-
tor can add or remove permissions of an application via the user interface of the security
application.
The applications are also given priority in order to ensure that the flow rule installation
of different applications do not interfere with one another. The priorities are stored together
with the permissions and are taken in to account while allowing applications to perform vari-
ous operations. A sample of the permission system is given in the Table 1.
Before granting the subscribed permission for an app, the security application checks if
it interferes with the operation of another application. The permission will only be granted
if the operation of a higher priority application is not affected by it. The applications are
allowed to ask the security app for a time slot for which to secure their operation, which will
be stored by the security app along with the permission set data. If there is a new request
for a permission before the expiration of the timeslot, the permission will be granted only if
it was requested by a higher priority application. The working of the security application is
given in Figure 1.
The security application also allows other applications to request for various permissions
and to know the permissions that are currently granted to them.
Table 1. Sample permissions set.
Application ID Allotted permission Priority
A read_topology X
read_all_flow
flow_mod_route
flow_mod_drop
flow_mod_modify_hdr
modify_all_flows
B pkt_in_event Y
flow_removed_event
error_event
topology_event
C flow_mod_modify_hdr Z
modify_all_flows
send_pkt_out
set_device_config
set_flow_priority
827

Figure 1. Security application.
5 CONCLUSION
The SDN architecture encourages the deployment of third party applications, but it also
introduces the additional task of ensuring which of these are genuine. This paper proposes a
security framework to safeguard the hybrid model of SDN by preventing malicious applica-
tions from taking control over the network. The hybrid model is an interesting attempt to
balance the controller load and once the various security issues are solved, it can be success-
fully used practically.
REFERENCES
Othman, O.M. & K. Okamura (2013a). Hybrid control model for flow-based networks. in the inter-
national conference COMPSAC 2013 – The First IEEE International Workshop on Future Internet
Technologies, Kyoto, Japan.
Othman, O.M. & K. Okamura (2013b). Securing distributed control of software defined networks,. Int.
J. Comput. Sci. Netw. Security 13, 5–14.
Porras, P., S. Shin, V.Y.M.F.M.T. &. G. Gu (2012). A security enforcement kernel for open flow net-
works. in Proceedings of the 1st workshop on Hot topics in SDN. ACM, 121–126.
Sandra Scott-Hayward, Member, I.S.N. & I. Sakir Sezer, Member (2016). A survey of security in soft-
ware defined networks. IEEE COMMUNICATION SURVEYS & TUTORIALS. 18, 623–654.
Scott-Hayward, S., C.K. & S. Sezer (2014). Operation check point: Sdn application control. 22nd IEEE
International Conference on Network Protocols (ICNP). IEEE, 618–623.
Wen, X., Y. Chen, C.H.C.S. & Y. Wang. (2013). Towards a secure controller platform for open flow
applications. in Proceedings of the second ACM SIGCOMM workshop on Hot topics in software
defined networking. ACM, 171–172.
828

A human-intervened CAPTCHA (HI-CAPTCHA) for wireless

LAN-based classroom applications
Pradeep Giri, Paras & Rajeev Singh

G. B. Pant University of Agriculture and Technology, Uttarakhand, India
ABSTRACT: CAPTCHA stands for Completely Automatic Public Turing test to tell
Computers and Humans Apart. CAPTCHAs are meant to distinguish between humans and
software bots and are used to prevent unauthorized access to Internet resources by the bots.
A CAPTCHA is a type of challenge-response test, in which a test is generated by the com-
puter program that most humans can pass but computer programs cannot. In this paper, a
new CAPTCHA approach is proposed: Human-Intervened CAPTCHA (HI-CAPTCHA).
This HI-CAPTCHA strengthens security in Wireless LAN (WLAN)-based mobile applica-
tions and systems. For example, in WLAN-based classroom applications, it is used for the
identification of the bots as well as for that of genuine users. A genuine user is one who is
authorized to use the WLAN system and is responding from inside the classroom.
Keywords: CAPTCHA, HI-CAPTCHA, Turing test, Information security, Mobile-based

system
1 INTRODUCTION
CAPTCHA stands for Completely Automatic Public Turing test to tell Computers and
Humans Apart. It is a mechanism used to distinguish between human users and computer
programs (von Ahn et al., 2004). The CAPTCHA term was coined in 2000 by Luis von Ahn,
Manuel Blum, Nicholas J. Hopper (all of Carnegie Mellon University), and John Langford
(then of IBM) (von Ahn et al., 2003). It is encountered in one form or another while using
different services, such as Gmail and PayPal, or online banking accounts (Baird & Popat,
2002; Datta et al., 2009). CAPTCHA is basically a challenge-response test that most humans
can pass but computer programs cannot. The rule of thumb of CAPTCHA is that it should
be solved easily by humans but not by software bots. CAPTCHA uses a reverse Turing test
mechanism (Coates et al., 2001). It has the following specifications:
• The judge is a machine rather than a human;
• The goal is that all human users will be recognized and can pass the test whereas no com-
puter program will be able to pass the test.
Thus, CAPTCHA helps in preventing automated and Artificial Intelligence (AI) software
programs known as bots from conducting unlawful activities on web pages, or in stopping
spam attacks on mail accounts. The bots try to automatically register for a large number of
free accounts and use these accounts to send junk email messages or to slow down services
by repeatedly signing in to accounts and causing denial of service. CAPTCHA stops such
autonomous entries and activities by bots in websites or in password-protected accounts.
Therefore, many websites utilize CAPTCHA against web bots. Some of the applications
(Carnegie Mellon University, 2010) of the CAPTCHA are as follows:
• Preventing comment spam in blogs
• Protecting website registration
829

• Protecting email addresses from scrapers and search engine bots
• Strengthening online polls
• Preventing dictionary attacks and worms.
CAPTCHA utilizes the fact that it is easy for humans to extract the text from the distorted
image whereas it is relatively difficult for the bots to understand the text hidden inside the
distorted image. As AI is evolving CAPTCHAs are becoming smarter and more secure. They
not only identify bots, but also help in the machine-learning process. For example, Google’s
reCAPTCHA mechanism not only helps in bot identification but also helps in some other
tasks such as Google’s book digitalization program, where a reCAPTCHA is used for the
learning of unidentified words which are not identified by Optical Character Recognition
(OCR) (Azad & Jain, 2013). Thus, the human effort required to solve the CAPTCHA is uti-
lized for some creative work, that is, for machine-learning processes.
Using CAPTCHA is always a win-win situation. If the CAPTCHA is solved by the bot, a
hard AI problem is solved; if not, it can be used for stopping bots [3]. A modified CAPTCHA,
HI-CAPTCHA, is proposed in this paper, which can be used for the identification of bots
as well as for the identification of genuine users. The proposed HI-CAPTCHA system is
targeted toward enhancing security in mobile-based classroom systems such as attendance
systems and examination systems. These WLAN-based mobile classroom systems face one
major challenge: that of identifying an authentic user in the classroom (termed a genuine
user) from one that lies outside the classroom (termed an invalid user). The proposed method
tries to distinguish these users based upon their responses.
2 RELATED WORK
Several Internet-based CAPTCHA mechanisms that use text, graphics, image, or sound exist
and are used by researchers to prevent bots and enhance security. These may be categorized
broadly into: first, plain CAPTCHA that is used only for the purpose of detecting and dis-
tinguishing humans and bots; second, other CAPTCHA that meets the above purpose and
utilizes the human effort in solving some problems. This paper aims to utilize the human
efforts used in solving the CAPTCHA for authentication purposes.
reCAPTCHA (Figure 1) was developed by Luis von Ahn, Ben Maurer, Colin McMillen,
David Abraham and Manuel Blum at Carnegie Mellon University (CMU). It was acquired by
Google in September 2009 (von Ahn et al., 2009). They studied harnessing the time and energy
of the human being, which is required for solving CAPTCHA. Every time the CAPTCHA
is solved, the effort used is utilized in digitalizing books, annotating images, and building
machine-learning data sets. It performs multiple tasks, that is, it identifies a bot, tries to solve
AI problems, and helps in machine learning. For example, it helps in utilizing the human effort
for digitalizing books in the following manner. The user is provided with two texts: the first
text is recognized by OCR while the other one is not. The user has to enter both texts and thus
helps in the machine-learning process by entering the non-identified word to the database.
The proposed HI-CAPTCHA system is also designed to utilize human effort to authen-
ticate users in the WLAN environment and to differentiate between a bot and human. It is
also designed to differentiate between genuine and invalid users/humans. The latter category
of users is not permitted to use/work with the system.
Figure 1. Google’s reCAPTCHA (von Ahn et al., 2009).
830

3 PROPOSED HUMAN-INTERVENED CAPTCHA (HI-CAPTCHA)
3.1 Proposed HI-CAPTCHA system

Today, CAPTCHAs are very much more secure. Many websites use CAPTCHA for identify-
ing bots as well as genuine users. Several users interact with these websites and try to solve
the CAPTCHA logic. Thus, a lot of human effort is involved to solve the CAPTCHA. This
human effort may be utilized in an effective way, as done by Google reCAPTCHA.
The proposed HI-CAPTCHA not only differentiates between a bot and a human but
also differentiates genuine and invalid users/humans. A genuine user is one who is author-
ized, stays in the classroom, and is entitled to use the system. For example, in a mobile-
based classroom exam, a user with a mobile device sitting in the class is a genuine user,
whereas one who is answering from outside the class is an invalid user. Hence the user’s
answer to the proposed HI-CAPTCHA determines whether he is genuine or invalid. Thus,
the user’s effort to answer HI-CAPTCHA is utilized in an effective manner for authentica-
tion purposes. The entire HI-CAPTCHA system is shown in Figure 2. The administrator
and clients have their own mobile application interfaces to interact with the system. They
connect via the WLAN Access Point (AP) to the Spring Tool Suite (STS) webserver and
database.
In the proposed HI-CAPTCHA system, questions are generated by the administrator. The
generated questions are sent to the user/client. The answer to the question varies as per the
administrator’s state and other things that can be identified by a person inside the room.
For example, in the question, ‘What is the color of the administrator’s shirt?’, the answer to
the question is not fixed as the administrator can wear any color shirt on any given day. The
answer depends on the administrator’s shirt color on a given day. Similarly, in the question,
‘Is the administrator moving or standing?’, the answer changes according to the administra-
tor’s direction of movement. The system works with a local server, where the numbers of
users connected to the local server are fixed. The HI-CAPTCHA questions will not only
detect the bot but will also help in authenticating the genuine user in selective mobile-based
systems. For example, in e-polling, attendance-based systems, online exams, and so on.
3.2 HI-CAPTCHA question

A human-intervened CAPTCHA system has customized questions. The questions are gener-
ated by the administrator. The questions are designed in such a way that the answer always
varies. The answer always depends on the current time, situation and location. The variable
answers to the questions provide additional security against bots and invalid users. This sys-
tem can be used for identifying bots and invalid users in small areas and for performing mul-
tiple tasks.
Figure 2. Block diagram of human-intervened CAPTCHA.
831

Some of the questions* and probable answers of the HI-CAPTCHA are:
• Is the administrator moving or standing?
Probable answer is either moving or standing.
The answer to the question varies according to the movement of the administrator.
• What is the color of the shirt the administrator is wearing?
Probable answer is red, green, blue, black etc. The answer to the question depends on the
color of the shirt the administrator is wearing.
• In which direction is the administrator standing?
Probable answer is east or west or south or north**. The answer to the question varies
according to the location, time and situation of the administrator.
4 HI-CAPTCHA RESULTS
The proposed HI-CAPTCHA system is developed on the Android platform. The main soft-
ware used for the development of the proposed system are Pivotal tc Server, Spring Tool
Suite and Android Studio 2.0. It is assumed that the user outside the room/class is not able
to see/perceive things inside the room/class. In HI-CAPTCHA, the administrator (teacher)
selects one question from the list of questions. The questions are designed in such a way that
the answer to the questions vary according to the administrator’s requirement; for example,
‘What is the color of the administrator’s shirt?’ The answer changes according to the color of
administrator’s shirt and the administrator can wear any color shirt. Hence, the answer is not
fixed and bots and outsiders are not able to answer the questions. Thus, the system remains
secure against bot attack and users positioned outside the room.
The proposed HI-CAPTCHA system is designed to differentiate between the genuine user and
the invalid user. For example, in mobile-based examination systems all users outside the class are
invalid users. A genuine user is one who is inside the classroom and authorized to use the system.
In the prototype of the system implemented (Figure 3a), the administrator has three
options to select: automated CAPTCHA, HI-CAPTCHA or an administrator task. If only
bots are to be restricted, plain automated CAPTCHA is selected. If invalid users are to be
restricted, HI-CAPTCHA is selected. The ‘administrator task’ option is used for chang-
ing passwords, registering users, and so on. Once HI-CAPTCHA is selected, other screens
(Figures 3b and 3c) are flashed up that ask for the setting up of questions and their answers
by the administrator in the current session. The duration in which a client’s response is
required is also selected (Figure 3d) depending upon network characteristics such as delay or
bandwidth. After finalizing the questions, answers and duration, the administrator sends the
question to the client to answer (Figure 3e). The client view of the HI-CAPTCHA is shown
in Figure 3f. It is used by the client to answer the administrator’s question correctly.
During the system run, all users who were in the class are able to see the administrator and
hence are able to answer the questions. Users outside the classroom were not able to see the
administrator and therefore they were not able to answer the questions and, therefore, were
identified as invalid users.
The proposed HI-CAPTCHA system works very well and detects bots and invalid users.
During the test, users outside the classroom were identified as invalid users. In Figure 3f, the
question selected by the administrator is: ‘In which direction is the administrator standing?’
The answer to the question depends on the direction of the administrator. All the valid users
in the class were able to see the administrator so they were able to answer correctly, while the
invalid user was not able to answer correctly.
*The HI-CAPTCHA system permits the administrator to design new and different questions as per
needs.
**The direction coordinates for this question may be decided by the instructor in the class and the stu-
dents briefed in advance.
832

Figure 3. HI-CAPTCHA test case.
HI-CAPTCHA is used for the identification of genuine users. A genuine user in the system is
one who is authorized to use the system. For example, in the mobile-based online examination sys-
tem, all the users inside the class are genuine users, while the users outside the class are not genuine.
The following test case is conducted to prove the result of the HI-CAPTCHA:
• The administrator selects the HI-CAPTCHA to authenticate the genuine user.
• A question selected by administrator is ‘Is the administrator moving or standing?’
• Answer to the question is based on the current time, situation or location of the adminis-
trator—administrator is standing.
• The time selected by administrator is 1 minute. This is the maximum time the user has to
answer the question.
All users inside the class answered correctly, as they can see the administrator. But a user outside
the classroom was not able to answer correctly because he was not able to see the movement of the
administrator. Hence, the administrator was able to identify the genuine users. Figure 3 shows the
diagrammatical view of the HI-CAPTCHA test case, for identifying a genuine user.
All the users inside the class were genuine users as they were authorized to use the pro-
posed system. However, a case may arise in which a user inside the class is not connected to
the server. In such a case, the user is not able to utilize the services of the system and will be
considered as an unauthorized user. There are a few situations in which this may occur:
833

• The application may be not installed on the device. The user has to install the HI-Client
CAPTCHA application.
• The device is not configured properly to use the system.
If the user is not able to connect to the server, the following condition may be checked:
• The user has to sign up to the HI-Client CAPTCHA application.
• The user inside the class has to log into the client HI-CAPTCHA application.
• If he/she logs in successfully, he/she may continue. But if the user fails to log in, the user
has to check the settings of the mobile, that is, check Wi-Fi settings, IP settings, and so
on.
In this case, when the user is in the classroom he must be treated as a genuine user. He may
ask the administrator to resend the question following proper connection with the server
after having been treated as not genuine/unauthorized.
5 CONCLUSION
Consistency of style is very important. Note the spacing, punctuation and capitals in all of
the examples above.
CAPTCHA is a very effective way for stopping bots and reducing spam. CAPTCHA keeps
web data secure from intruders. Almost every website contains CAPTCHA in one form or
other. Every ‘Sign in and sign up’ or form submission over the Internet contains CAPTCHA.
As AI is evolving, the need to develop new and advanced forms of CAPTCHA has arrived.
Google’s reCAPTCHA is an example of an advanced CAPTCHA. It not only detects soft-
ware bots, but also helps in the machine-learning process. In the same way, the proposed
CAPTCHA system is also designed for the multi-tasking environment. The proposed HI-
CAPTCHA not only differentiates between a bot and a human but also differentiates between
genuine and invalid users/humans. The proposed HI-CAPTCHA system can be designed for
WLAN-based mobile attendance systems, e-polling systems or for generating the details of
users. It works very well in a local server environment. In future, it may also be tested along
with new CAPTCHA variants such as NO CAPTCHA and reCAPTCHA (Google, 2017).
REFERENCES
Azad, S. & Jain, K. (2013). CAPTCHA: Attacks and weaknesses against OCR technology. Global Jour-
nal of Computer Science and Technology, 13(3).
Baird, H.S. & Popat, K. (2002). Human interactive proofs and document image analysis. In D. Lopresti,
J. Hu & R. Kashi (Eds.), Document analysis systems V. DAS 2002. Lecture notes in computer science
(Vol. 2423, pp. 507–518). Berlin, Germany: Springer.
Carnegie Mellon University. (2010). CAPTCHA: Telling humans and computers apart automatically.
Retrieved from http://www.captcha.net.
Coates, A.L., Baird, H.S. & Faternan, R.J. (2001). Pessimal print: A reverse Turing test. In Proceedings
of 6th International Conference on Document Analysis and Recognition, Seattle, WA (pp. 1154–1158).
New York, NY: IEEE.
Datta, R., Jia, L. & Wang, J.Z. (2009). Exploiting the human-machine gap in image recognition for
designing CAPTCHAs. IEEE Transactions on Information Forensics and Security, 4(3), 504–518.
Google. (2017). Introducing the new reCaptcha! Retrieved from https://www.google.com/recaptcha/
intro/index.html.
von Ahn, L., Blum, M., Hopper, N.J. & Langford, J. (2003). CAPTCHA: Using hard AI problems for
security. In E. Biham (Ed.), Advances in cryptology—EUROCRYPT 2003. Lecture notes in computer
science (Vol. 2656, pp. 294–311). Berlin, Germany: Springer.
von Ahn, L., Blum, M. & Langford, J. (2004). Telling humans and computers apart automatically. Com-
munications of the ACM, 47(2), 57–60.
von Ahn, L., Maurer, B., McMillen, C., Abraham, D. & Blum, M. (2009). reCAPTCHA: Human-based
character recognition via web security measures. Science, 321(5895), 1465–1468.
834

Semantic identification and representation of Malayalam sentence

using SVM
Shabina Bhaskar, T.M. Thasleema & R. Rajesh

Department of Computer Science, Central University of Kerala, Kasaragod, India
ABSTRACT: This paper introduces Semantic Role Labeling (SRL) method for Semantic
Identification (SI) and Conceptual Graph techniques (CG) for semantic representation. In
Semantic Role Labeling, the semantic roles have been identified as agent, patient etc using
the Karaka theory. The performance of SRL is calculated using Yet Another Chunk Annota-
tor, a Support Vector Machine based algorithm which has shown significant improvement
over earlier methods. The semantic representation introduced here is a directed graph which
shows the relation between concepts and semantic roles.
1 INTRODUCTION
Semantics is the study of linguistic utterance. It refers to the sentence level meaning which is
context independent and purely linguistic. The semantic approaches introduced has manifold
applications in various NLP areas such as question answering system, machine translation
system, text summarization system etc. Semantic identification and representation is a very
important issue in natural language processing.
A sentence is represented by predicates and its corresponding arguments. Predicate repre-
sents an event and abstract roles are semantic roles that the arguments of a predicate can take
in an event. For meaning identification and representation, we must identify the predicates
and the abstract roles in a sentence. For high level understanding many question types need
to be dealt with such as who did, what, to whom. Semantic Role Labeling, a shallow semantic
parsing approach tries to find answers to these questions (Jurasfky & Martin 2002). Thus it
takes the preliminary steps in extracting meaning from a sentence by giving generic labels or
roles to the tokens of the text (Guilda. & Jurasky 2002).
The Semantic Role labeling is implemented (Kadri et al. 2003) using SVM classifier. The
results evaluated using both hand-corrected TreeBank syntactic parses, and actual parses
from the Charniak parser shows a precision and recall rate of 75.8% and 71.4% for this
work. The semantic role labeling based on syntactic chunks is presented in (Kadri & Wayn
2003) shows some inspirational results with a precision and recall rate of 76.8% and 73.2%
respectively.
Semantic role labeling methods in Malayalam is presented in (Dhanya, P.M 2010, Jisha &
Satheesh 2016) and roles identified are used for general concept understanding and concept
representation (Radhika & Reghuraj 2009). Plagiarism detection in Malayalam document
based on extracting the Semantic roles and computing their similarity (Sindhu & Suman
mary 2015) is another work related to Semantic role labeling.
The conceptual graph representation proposed by Sowa (Sowa 2008) express the mean-
ing of a sentence in logically precise, humanly readable and computationally tractable form.
In CG, we have to contemplate about the concepts and their corresponding semantic rela-
tions. This technique has been applied to many real life objects including text (Paola et al.
2008). The CG is used in relation extraction, information extraction and many other concept
extraction techniques (Montes et al. 2001). Fact Extraction is a part of more general prob-
lem in knowledge extraction from text (Yi Wan et al. 2014). The fact extraction from natural
835

language texts with conceptual graph is explained in (Bogatyrev 2017). For fact extraction,
the input data is in the form a text and the text is transformed to a set of conceptual graph
and the maximal number of conceptual graph is equal to the number of processed sentence.
The concepts are identified from this CG and then presented in the form of a concept lattice.
Concept lattice represents a matrix with relation on two sets of objects and their attributes.
The facts can be extracted by processing the input textual queries in concept lattice. Concept
and relationship extraction from unstructured text data plays a key role in meaning aware
computing paradigms (Anoop & Ashraf 2017).
Malayalam, a morphologically rich and agglutinative language contribute to the ardu-
ousness of semantic identification. In Malayalam, the root or the stem word by itself gets
inflected to change its meaning, or to combine the word with other words. Malayalam being
a relatively free word order language, semantic role identification becomes a strenuous activ-
ity. The feature of inflection and agglutination nature of Malayalam be partly responsible
for computer-based language processing a challenging task. For semantic identification in
Malayalam, we have to associate different roles connected with various verbs, which is known
as thematic role analysis or case role (Karaka) analysis. Thus karaka analysis plays a piv-
otal role in finding the relations between verbs and nouns in a sentence and we distinguish
abstract semantic roles or thematic roles such as agent, patient etc. According to Paninian
perspective there are four levels in the understanding process of a sentence (Bharati et al.).
They are surface level (uttered sentence), vibhakthi level, karaka level and semantic level. The
karaka level has relationship to semantics on one side and to syntax on the other side. Karaka
relation can be identified from post position markers after noun or surface case ending of
noun. These markers and case endings are called vibhakthi (Manu & Reghu 2012). In Malay-
alam karaka relations are analyzed from vibhakthi and post position markers (Archana et al.
2015). But for a sentence level meaning identification we have to consider the semantic roles
associated with adjectives and adverbs also. The semantic roles of adjectives, adverbs and
postposition markers are also identified in this work.
After Semantic role identification, it will be pursued with the meaning representation.
The intermediate representation presented here is a basic level meaning representation of
a sentence. The intermediate representation gives us information about semantic roles of
nouns, adjectives, adverbs and the verbs in a sentence. And through the introduction of the
conceptual graph for meaning representation, this graph manifests how the verb is related to
noun, adjectives and adverbs in a sentence.
In section 2 we discussed about the methodologies used in this work. And section 3 and 4
discussed about the results and conclusion.
2 METHODOLOGY
Overall system architecture is given in Figure 1. For syntactic level processing the main steps
are tokenization, POS tagging and Morphological analysis. In semantic side, the main steps
are semantic role labeling and intermediate representation.
2.1 Tokenization
Tokenization is the task of chopping a character sequence into pieces, called tokens, perhaps
at the same time throwing certain characters, such as punctuation. For tokenization we devel-
oped an algorithm and for compound word splitting SVM machine learning techniques used.
It classifies the tokens into two groups compound words and simple words.
2.2 POS tagging

The process of assigning part of speech for every word in a given sentence according to the
context is called as part of speech tagging. SVM classifier is used for POS tagging and the
system is trained with annotated corpora. For annotation we used BIS tagset.
836

2.3 Morphological analysis

Morphological analysis is the process of recognizing the root stem, and the categorical infor-
mation of the items that may accompany the root or the stem. Here for Morphological analy-
sis, Paradigm based approach is used and we developed Paradigms for nouns and verbs.
This Paradigms includes high level information such as vibhakthi and verb forms other than
categorical information.
2.4 Semantic role labeling

Semantic Role Labeling is the task of assignment of the semantic roles to the constituents of
the sentence. The semantic roles such as agent, patient, beneficiary, experiencer, instruments,
recipients and locative are identified in this work. The Table 1 gives you information about
the semantic roles used in this work.
2.4.1 Description about semantic role labeling method

In Malayalam Karaka relation express the syntactic and semantic relations exists between
words in a sentence. This karaka relation and vibhakthi helps us to identify the semantic roles
in a sentence. See Table 2.
But in some situation we cannot identify the semantic roles with the help of karaka rela-
tions alone. In such cases we have to add some additional semantic information about noun
and verb. In this work we have created a lexicon of noun and verb in which we grouped the
nouns and verbs which have same semantic property. For example, the words like dance, sing
have the same semantic property and it is grouped as artistic performance verbs. The noun
class and verb class used for this work is given below.
Verb class
1. Artistic Performance verb ( )
2. Attach Verb ( , , )
3. Beneficiary Verb ( , )
4. Causative Verb ( , , )
5. Emotional verb ( , )
837

)
)
Table 1. Semantic roles.

) Semantic roles Definition
Agent Semantic role of a person or thing who is the doer of an event.

Patient Semantic role that is usually the surface object of the verb in a sentence.
Beneficiary Semantic role of a referent which is advantaged or disadvantaged by an event.
Experiencer Semantic role of an entity that receives sensory or emotional input
Instrument Semantic role of an inanimate thing that an agent uses to implement an event.
It is the stimulus or immediate physical cause of an event.
Recipients Semantic role of a person or thing one who receives the goal of an action.
Locative Semantic role which identifies the location or spatial orientation of a state or action
)
Table 2. Relation between karaka and roles.
Karakas Vibhakthi (case role) Semantic role
Karthu Nideshika Agent

Karma Prathigrahika Patient
Karana Prayogika Instrument
Kaarana Prayogika Instrument
Sakshi Samyogika Experiencer
Swami Udheshika Beneficiary
Adhikarana Adharika Locative
6. Destroy Verb ( , )
7. Vehicle Motion Verb ( , )
8. Weather Verb ( )
Noun class
1. Animals ( )
2. Audible only ( )
3. Birds ( )
4. Buildings ( )
5. Collective humans ( )
6. Concepts ( )
7. Positions ( )
8. Person names ( )
9. Non-living ( )
Next we discussed some examples about how we can identify semantic roles with the help
of karaka, vibhakthi and semantic property.
1. Nirdeshika Nirdeshika + verb (emotional verb) The semantic role is
experiencer.
Here the sentence is subject + object + verb form and if the verb is causative verb then
the noun with nirdeshika vibakthi take the role as Agent. shows prathigrahika
vibakthi and the role is patient.
2. Udeshika Udeshika + verb Here the verb is beneficiary verb then role
is beneficiary or recipient.
3. Nirdeshika + nirdeshika + verb
Here the first noun has the semantic property person name and second is fruit name, in
that case person takes the role of agent and the other takes patient.
838

Here we check the semantic property of noun both are non-human. So we take the order
the first become agent and second become patient.
4. Nirdeshika + prathigrahika Here the role is agent + patient
5. Nirdeshika + udeshika
Verb is beneficiary so semantic role is beneficiary.
6. Nirdeshika + udeshika + nirdeshika Role is agent +
beneficiary + patient.
2.4.2 YamCha for semantic role labeling

YamCha is Yet Another Multipurpose Chunk Annotator which is based on the SVM
machine learning algorithm, first introduced by Vapnik in 1995. It Use PKE/PKI, which
make the classification (chunking) speed faster than the original SVMs. Here the features
selected are words in a sentence, their POS tag and semantic roles. In this work we cre-
ated corpora in which each words and its semantic roles are annotated. Other than the
semantic roles of noun we have also identified the semantic roles for adjectives, adverbs
and verbs.
2.5 Intermediate representation

Intermediate representation is which gives us information about semantic roles, verbs, adjec-
tives and adverbs in a sentence. In this representation we grouped nouns, verbs, adjectives,
adverbs and postposition ma rkers separately.
Nouns
• AGENT 1
• AGENT2
• PATIENT1
• PATIENT2
• BENEFICIARY
• EXPERIENCER
Adjectives
• APPEARANCE
• COLOR
• MEASUREMENT
Adverbs
• FREQUENCY
• MANNER
Postpositions
• DESTINATION
• LOCATION
Verbs
• MAIN VERB
• SUB VERB
In this work the maximum number of verbs in a sentence is two, one is main verb and other
is sub verb. AGENT1, PATIENT1 are the roles for main verb and AGENT2, PATIENT2 are
the role for sub verb. The intermediate representation for a sentence with two verbs is given
below.
Input sentence:
839

Figure 2. Conceptual graph.
Figure 3. Semantic representation in the form of a directed graph.
Noun
AGENT1
AGENT2
PATIENT1 NIL
PATIENT2
Adjectives
NIL
Adverbs
NIL
Postpositions
NIL
Verbs
MAIN VERB
SUB VERB
2.6 Conceptual graph representation

Conceptual Graphs (CG) serve as an intermediate language for translating computer-ori-
ented formalisms to and from natural languages. They serve as a readable, but formal design
and specification language. CGs have been implemented in a variety of projects for infor-
mation retrieval, database design, expert systems, and natural language processing [19]. In
diagrammatic representation of CG rectangular box represents the concept node and oval
represents the semantic relation (semantic roles). CG are based upon the following general
form:
In this work we adopt the techniques of Conceptual Graph representation for meaning
representation and here the concepts-semantic relation is expressed in the form of a directed
graph. The concepts are represented by vertices and edges represents their corresponding
semantic relations. The main verb is considered as the root node and the child nodes are
nouns, adjectives and adverbs. But in a complex sentence, the main verb is root node and sub
verb is its right child node. Semantic representation of a sentence is
given in Figure 3.
An algorithm is introduced for semantic representation of sentence in the form a graph.
840

841

A total of 300 and 50 sentences are used for training and testing respectively. In testing,
out of sentences for SRL we took 40 simple sentence with word length from two to four and
10 complex sentence with word length five. The semantic roles corresponding to each word is
identified and it is transformed in to a directed graph representation.
3 RESULTS
In this experiment the sentence word length is minimum two and maximum five. We have
taken the complex sentence with maximum two verbs. In our study the evaluation is based on
the standard metrics such precision and recall.
The precision for a class is the number of true positives divided by the total number of ele-
ments labelled as belonging to the positive class (i.e. the sum of true positives and false posi-
tives, which are items incorrectly labelled as belonging to the class). Recall in this context is
defined as the number of true positives divided by the total number of elements that actually
belong to the positive class (i.e. the sum of true positives and false negatives). The precision,
recall rate corresponding to each role is given in Table 3.
The result of semantic role labeling and semantic representation in the form of graph is
given below
Input sentence:
Semantic roles:
AGENT1
APPEARANCE
PATIENT1
MAIN VERB
Graph
Table 3. Precision and recall for semantic roles.
Semantic roles Precision Recall
Agent1 81.2% 78.2%

Agent 2 69.0% 65.4%
Patient 1 80.0% 76.4%
Patient 2 75.0% 68.3%
Beneficiary 80% 76.8%
Experiencer 75.5% 60.6%
Instrument 70.6% 64.0%
Appearance 73.0% 68.1%
Colour 67.2% 60.0%
Measurements 60.6% 58.8%
Frequency 60.9% 56.8%
Manner 62.2% 58.0%
Location 66.4% 59.9%
Main verb 80.9% 76.9%
Sub verb 78.8% 73.3%
842

4 CONCLUSION
The SVM based classifier introduced here considered the Semantic Role Labeling (SRL) as a
semantic grouping problem of different words. The SRL identified the semantic roles associ-
ated with each words in a sentence. The Conceptual Graph presented here shows the relation
between predicates and arguments in a sentence. The CG representation depends on the
semantic role labeling. If the roles are identified correctly, then it will lead to more accurate
system. By introducing more semantic roles and by increasing the sentence word length we
can add improvements in the present study.
REFERENCES
Anoop, V.S. & Ashraf. 2017. Extracting Conceptual Relationships and Inducing Concept Lattices from
Unstructured Text. Journals of intelligent systems.
Archana, S.M. & Vahad, Naima. & Rekha, Thankappan. & Raseek. C. 2015. A Rule Based Question
Answering System in Malayalam corpus Using Vibhakthi and POS Tag Analysis. International Con-
ference on Emerging Trends in Engineering, Science and Technology (ICETEST – 2015).
Bharati, Akshar. & Vineeth, Chaithanya. & Sangal, Rajeev. Natural Language Processing: A Paninian
Perspective. Prentice-Hall of India, New Delhi.
Bogatyrev, Mikhail. 2017. Fact extraction from natural language with conceptual modelling, Springer
international Publishing AG.
Dhanya, P.M. 2010. Semantic Role labeling Methods: A Comparative Study. In: proceedings of National
Conference on Human Computer Interaction and Image Processing. Vidya Academy of Science &
Technology.
Guilda & Jurasky, Daniel. 2002. Automatic Labelling of semantic roles. 245–288.
Gildea. & Hockenmaier. 2003. Identifying Semantic Roles Using Combinatory Categorical Grammar.
2003. In Proceedings of the 2003 conference on Empirical methods in natural language processing.
Association for Computational Linguistics, Philadelphia, USA, pp. 57–64, 2002.
Kadri, Hacioglu. & Pradhan, Sameer. & Wayne, Ward. & Martin, James. & Jurasfky, Daniel. 2003. Shal-
low Semantic Parsing Using Support Vector Machines. CSLR Tech. Report.
Kadri, Hacioglu. & Wayne, Ward. 2003. Target word detection and semantic role chunking using sup-
port vector machines. In Proceedings of HLT-NAACL 2003.
Jish, Jayan. P. & Kumar, Satheesh. 2016. Semantic role labeling for Malayalam. IJCTA. pp. 4725–4731
© International Science Press.
Jurasfky, Daniel. & Martin, James. 2002. An introduction to Natural Language Processing: Computa-
tional Linguistics and Speech Recognition. Pearson Education.
Manu, Madhavan. & Reghu, Raj. P.C. 2012. Application of Karaka Relations in Natural Language
Generation. In Proc. of National Conference on Indian Language Computing, CUSAT, Kerala.
Montes, Y. Gomez. M. & Gelbukh, A. & Lopez-Lopez & Baeza-Yates. Text mining with conceptual
graphs. 2001. IEEE International Conference on Systems, Man and Cybernetics. e-Systems and e-Man
for Cybernetics in Cyberspace (Cat. No. 01CH37236) Year: 2001, Volume: 2.
Paola, Velardi. & Maria, Teresa. Pazienza. & Mario De’ Giovanetti. 1998. Conceptual graphs for the
analysis and generation of sentence. IBM journal of Research and Development.
Radhika, K.T. & Reghuraj, P.C. 2009. Semantic role extraction and General concept understanding in
Malayalam using Paninian Grammar. International Journal of Engineering Research and Develop-
ment, vol 9, issue 3.
Sindhu, L. & Suman Mary, Idicula. 2015. SRL based Plagiarism Detection system for Malayalam Doc-
uments. International Journal of Computer Sciences issues, vol 12, November 2015.
Sowa. John. 2008. Conceptual Graph. Chapter 5 of the Handbook of Knowledge Representation, ed.
by F. van Harmelen & V. Lifschitz, & Porter. B, Elsevier, 2008, pp. 213–237.
Stephen, chu. & Branko, Cesnik. 2001. Knowledge representation and retrieval using conceptual graphs
and free text document self-organisation techniques, International Journal of Medical Informatics 62
(2001) 121–133.
Yi, Wan. & Tingting, He & Xinhui, Tu. 2014. Conceptual graph based text classification. IEEE Interna-
tional Conference on Progress in Informatics and Computing 2014.
843

A survey on relation extraction methodologies from

unstructured text
S.K. Bhaskaran & P.C. Rafeeque

Department of Computer Science and Engineering, Government Engineering College, Palakkad, India
ABSTRACT: The web contains very large amount of unstructured text. The process of
converting unstructured text to structured with semantic information annotation can provide
useful summaries for both humans and machines. The semantic relation is one of the most
important parts of semantic information. Hence, extracting semantic relations held between
entities in a text is important in many natural language understanding applications like ques-
tion answering, conversational agents, summarization. There are several proposed methods
in this area. Hand built patterns, bootstrapping methods, supervised methods, unsupervised
methods, and Distant Supervision (DS) are examples. The purpose of this work is to review
various methods used for Relation Extraction (RE). For each approach, respective motiva-
tion is discussed and the merits and demerits compared. Discussion about various datasets is
also included in this paper.
1 INTRODUCTION
1.1 Information extraction

The main goal of an Information Extraction (IE) task is to provide machine-readable summa-
ries (ie, acquiring structured information from unstructured text). Normally, documents on
the web are unstructured and are not understandable by a machine. Thus, IE aims to provide
machine-readable summaries. Usually, IE means extracting information from text; sometimes
called text analytics commercially. Now, what type of information can be extracted? It can
extract entities, the relationship between entities and it can also figure out larger events that
are taking place. Normally, interested entity types are the names of people, organizations,
locations, times, dates, prices or, sometimes, genes, proteins, diseases and medicines. Some of
the relations between entities are located_in, employed_by, part_of, married_to etc.
1.2 Relation extraction

The amount of unstructured electronic text on the web is very large. So, understanding this text
is a complicated task for humans. A popular idea is to transform this unstructured text into
structured by annotating semantic information. But human annotation is impossible because
of the large volume and heterogeneity of data. It is possible for an agent (or computer) to do all
these annotations. So, the agent (or computer) needs to understand how to recognize whether a
piece of text holds some semantic property or not. Hence, extraction of the semantic relations
between entities is an important task. The Relation Extraction (RE) task aims to recognize
and classify relations between pairs of entities found in a text. For instance, the RE system will
extract CEO-of (Jobs, Apple) from the sentence “Jobs is the CEO of Apple”.
Traditional methods for RE handle the task as a pipelined approach. That is, first extract
the entities and then recognize the relationship between those entities. This is particularly easy
to handle and will assure the flexibility in each subtask. But it has some drawbacks; specifically
the relevance of each subtask is neglected and we have to model both subtasks separately. Also,
it results in error propagation and redundant information. To tackle the above problems, it is
845

useful to think about a joint learning framework. Here, a single model is used to extract entities
together with their relations. It provides an effective integration of both subtasks. However, exist-
ing joint extraction methods are feature-based which may require complicated feature engineer-
ing and have great dependency on other NLP toolkits. It also produces redundant information.
A neural network-based approach can reduce the manual work in feature extraction. Intro-
duction of a deep learning framework to RE has made a significant performance improve-
ment to traditional methods. In this review, we begin with problem formulation, and we
discuss various methods proposed for RE task. We will highlight the merits and demerits of
each method and if it is useful to choose a better methodology. It will also help us to make
new proposals for existing works.
Organization: The rest of this paper is organized as follows. Section 2 formulates the prob-
lem and explain about challenges of the RE problem. It also addresses some applications in
NLP. Section 3 discusses various pipelined methods, jointly extracting methods and the end-
to-end tagging models. Section 4 discusses future enhancements to existing RE methods. The
conclusion is given in section 5.
2 FORMULATION OF THE PROBLEM
The inputs to an RE system are a POS-tagged corpus, C, a Knowledge Base (KB), K, and a
predefined relation type set, R. An entity mention is a token span, denoted as e and r(e1,e2)
is the relation (binary) between entities e1 and e2. The manual labeling of training corpora
is expensive in terms of time. The method of Distant Supervision (DS) can overcome this
problem and it is the reason why DS has achieved significant importance in RE tasks.
DS generates training data by automatic alignment of text with a KB. It can jointly extract
entities together with their relations with minimal or no human supervision. Hence, KB is
also an input to the RE system.
PROBLEM DEFINITION: Given a POS-tagged corpus, C, a KB, K, and predefined rela-
tion type set, R, the RE task aims to 1) detect entities from C, 2) generate training data D with
KB, K, and 3) estimate a relation type r which belongs to R U {None}.
Current RE tasks with DS face the following limitations when handling a joint extraction task:
• Domain Restriction: Most methods (Mintz et al., 2009; Takamatsu et al., 2012) heavily
rely on pre-trained named entity recognizers which are typically designed for general types
such as person or organization. We need further manual work to deal with domain specific
names.
• Error propagation: Error can be propagated from upper components to lower components
and dependencies among tasks are ignored in most existing methods.
• Domain-independent systems: A major challenge is to design a domain-independent sys-
tem. Most existing methods are domain dependent.
• Label Noise: Mapping of relations in the text with KB relations may produce false labels
in training corpora. It may cause uncertainty in DS and thereby results in inaccurate
models.
RE is the extraction of semantic relations from unstructured text. Once extracted, such
structured information is used in many ways. For example, as primitives in IE, building
extending KBs and ontologies, question answering systems, semantic search, machine read-
ing, knowledge harvesting, paraphrasing, and building thesauri.
3 RELATION EXTRACTION METHODS
In this section, we briefly explain three methods employed for the problem of RE. The two
main frameworks used for this are the pipelined framework and the joint learning frame-
work. Also, some neural network-based methods for extracting entities and relations are
explained for understanding the tagging approaches used in this area.
846

3.1 Pipelined methods
Traditional methods for RE handle the task in a pipelined manner. That is, they perform
Named Entity Recognition (NER) first and then classify the relations. The NER task is
similar to POS tagging and sequence classifiers such as Hidden Markov Models (HMM)
and Conditional Random Fields (CRF) which are used for this task. Recently, Long Short
Term Memory (LSTM) for NER has been achieved significant performance improvement
over sequence classifiers. Now, we move on to the second subtask, namely relation classifica-
tion. Existing methods for relation classification use feature-based methods as well as neural
network-based methods. Let us discuss some of the pipelined methods.
Mintz et al. (2009) proposed a domain-independent model for RE which uses DS as an
alternative paradigm for existing supervised, unsupervised and semi-supervised or bootstrap-
ping methods. DS performs automatic labeling of relations in text and relations in KB. It
uses a multi-class logistic classifier for extraction. The method uses relations and relation
instances from Freebase, which is an online database of structured semantic data. For a Free-
base relation with a pair of entities, all sentences containing those entities are retrieved from
the text. It combines advantages of both supervised IE and unsupervised IE.
In this method, textual features like lexical features, syntactic features, and name entity
tag features are extracted and trained by the classifier using these features. Unlike a corpus-
specific method, the main advantage of this method is that it allows us to extract evidence
for a relation from many documents and thereby avoids domain dependence. Another advan-
tage of this approach is that it does not require a labeled corpus. Since it is a feature-based
approach, the main limitation is that it needs feature extraction which requires manual work.
Moreover, uncertainty in DS is another problem since it exploits only one specific kind of
indirect supervision knowledge and ignores very useful supervision knowledge. To overcome
this limitation, we can think about global DS which will effectively reduce uncertainty by
including additional knowledge.
In the method by Takamatsu et al., (2012), when a sentence refers to an entity pair in a KB,
then that sentence is heuristically labeled by DS with the corresponding relation in the KB.
However, this labeling will result in wrong labels and thereby poor extraction performance.
The model effectively reduces incorrect labeling by predicting whether assigned labels are
correct or not. This is done by the model’s hidden variables. The advantage is that it reduces
wrong labels but it does not address inter-dependencies.
Surdeanu et al. (2012), introduced a challenging learning scenario where the relation held
by a pair of entities is unknown for each sentence. The method uses a multi-instance multi-
label learning (MIML) framework, which can jointly model all the entities and labels by a
graphical model. The model assumes that a relationship between two entities has exactly
one label, but the entities can have multiple labels. Two separate feature sets are used for the
entity classifier and the relation classifier respectively. One of the merits of this work is it also
reduces incorrect labels. The limitation of this approach is that it needs feature extraction for
both subtasks.
3.2 Jointly extracting methods

In a joint extraction framework, entities are extracted together with relations between pair
of entities.
The method by Li and Ji (2014), is a joint extraction method with DS. It exploits global fea-
tures in the joint search space. The joint framework is an incremental one and uses a structured
perceptron with the advantage of efficient beam-search. A segment-based decoding is another
specialty of this model which is based on the idea of a semi-Markov chain and is employed
as an opposition to traditional token-based tagging schemes. This method has developed a
number of new and effective global features to capture the inter-dependency among entities
and relations. In this method, segment-based features are considered along with the global
features. Segment-based features are based on the entire mention instead of individual tokens.
They are gazetteer features, word case features (case information about all tokens contained),
847

contextual features (neighbor unigrams and bigrams), and parsing features such as the phrase
label of common ancestor (NP), the depth of the common ancestor, whether the segment
matches a base phrase (true) or is a suffix of a base phrase and the head word of the segment.
Global features are dynamically created during the search, and they capture long distance
dependencies. One merit of this work is that it captures the inter-dependency among entities
and relations. Uncertainty in DS is again the drawback of this method.
Hoffmann et al. (2011), proposed a knowledge-based weak supervision approach. The
framework is known as MultiR. It uses heuristic labeling and multi-instance learning algo-
rithms for coping with noisy data. MultiR is a probabilistic, graphical model which employs
multi-instance learning technique and handles overlapping relations. It also results in accu-
rate sentence-level predictions, decoded individual sentences, and corpus-level extractions.
Here, RE is modeled as a MIML problem, but it learns to use a perceptron algorithm. Also,
it introduced at least one deterministic decision rather than using a relation classifier.
CoType is an embedding based framework proposed by Ren et al. (2017). It is a novel
distant supervision framework that can extract entity mentions and relations jointly with min-
imum linguistic supervision. For detecting entities using DS, this work developed a context-
agnostic algorithm. An overview of framework is as follows: The algorithm is first run over
a POS-tagged corpus which is the input to the CoType. It generates entity relation mentions
and extracts text features for each relation. Thus, DS is used for generating training corpora. A
joint embedding objective is also formulated for modeling type association, mention feature-
co-occurrence, and entity-relation cross-constraints. Experiments on three public datasets have
demonstrated that CoType improves extraction problem performance significantly. Advan-
tages: 1) Domain-independent; 2) Modeling of type association and inter-dependencies. The
drawback of this work is that it does nothing for reducing false labels in the training data.
3.3 End-to-end tagging models

End-to-end tagging models use several deep learning architectures for the extraction problem
into a tagging task.
Lample et al. (2016) proposed a neural network-based approach for state-of-art NER.
There are two neural networks: one is LSTMs and CRFs based and a transition-based
approach is used by the second neural network for constructing and labeling segments. The
tagging scheme used here is IOBES (Inside, Outside, Beginning, End, Single). An advantage
of this method is that the features extracted are not language specific.
Another method by Zheng et al. (2017) is a joint extraction method based on a tagging
scheme. It uses an IOBES tagging scheme and transforms the extraction problem into a
tagging task. An LSTM based framework is employed and results show that tagging based
methods are more efficient than other pipelined and jointly extracting methods. NER results
are extracted using CoType which uses DS. The main advantage is that it ignores complicated
feature engineering. But there is a shortcoming for the identification overlapping relations to
settle the triplet overlapping problem.
3.4 Discussion on datasets

The above discussed methods use publicly available datasets such as New York Times articles
(NYT) and Wikipedia articles (Wiki-KBP). Bioinfer is a commonly used dataset from the
medical field. But these are not annotated. An effective and efficient way for annotation is DS
which is the most accepted approach.
Approaches in Hoffmann et al. (2011), Mintz et al. (2009) and Surdeanu et al. (2012) use
Wikipedia datasets and Freebase as the KB. The dataset from Wikipedia is an easy sce-
nario for DS with Freebase because the facts in Freebase are partly derived from Wikipedia.
Experiments on a CoType framework by Ren et al. (2017) used three public datasets, NYT,
Wiki-KBP, and Bioinfer which contain biomedical paper abstracts. Bioinfer contains several
overlapping relations. Out of these three sets, NYT is used with Freebase as the KB in the
method by Zheng et al. (2016).
848

Table 1. Overview of compared methods and datasets used.
Proposed methods
Approach Model Contributions Domain/Dataset
Pipelined Mintz et al., DS-logistic, allow corpora of any size Wikipedia with Freebase
2009 (9)
Takamatsu et al., DS, reducing wrong labels Wikipedia with Freebase
2012 (14)
Surdeanu et al., DS, multi-instance NYT, KBP and Freebase
2012 (13) multi-learning algorithms
Joint Li and Ji, DS, exploits global features ACE 2005
2014 (7)
Hoffmann et al., MIML, sentence-level predictions, Wikipedia with Freebase
2011 (5) reducing false labels
Ren et al., DS, CoType, modeling type association, NYT, Wiki-KBP, Bioinfer
2017 (10) mention feature-co-occurrence, and
entity-relation cross-constraints
Tagging Lample et al., LSTM-CRF, transition-based CoNIL 2002, CoNIL 2003
Based 2016 (6) approach, IOBES tagging
Zheng et al., DS, a novel tagging scheme, NYT and Freebase
2017 (17) Bi-LSTM-LSTM
In the method by Li and Ji (2014) their experiments were done on an ACE 2005 data-
set. It contains data from six different domains such as Newswire, Broadcast Conversation,
Broadcast News, Telephone Speech, Usenet Newsgroups and Weblogs. Different datasets
were used by Lample et al. (2016); CoNIL 2002 and CoNIL 2003 which contain independent
named entity labels for English, Spanish, German and Dutch. Since different methodologies
used different datasets, a general comparison is not possible. Contributions of the different
methodologies discussed and domains or datasets used by them are summarized in Table 1.
4 FUTURE DIRECTIONS
Many natural language understanding applications need competitively structured informa-

tion. So, RE is always important. Most of the proposed methods employ a joint extraction
task with DS. Enhancements in DS will fill the gaps in existing methods. One of the problems
with DS is uncertainty which may lead to lack of supervision. Introduction of a pseudo feed-
back idea to the existing framework may overcome this. Feature enrichment of the current
embedding framework is an interesting enhancement. Unlike the method used in Hoffmann
et al. (2011), it is possible to develop an end-to-multiple-end model for addressing triplet
overlapping problem. Modeling type correlation and performing type inference for test entity
and relation jointly will be other interesting future work in this area.
5 CONCLUSIONS
Semantic relations in text are the key meaning component for natural language applications.
There are several proposed approaches for the RE task and here we have given a brief summary
of some of them. So far, we have reviewed some important aspects of the entity RE problem
starting with the problem formulation, discussing the different challenges and applications
and finally culminating with a discussion of some important approaches. Various methods
of pipelined approaches, joint extraction framework and neural network approaches, which
use end-to-end tagging schemes have been discussed, and we can clearly say that end-to-end
849

models can outperform other models. Moreover, deep learning frameworks have replaced
most of the existing frameworks because of their popularity.
REFERENCES
Bollacker, K., Evans, C., Paritosh, P., Sturge, T. & Taylor, J. (2008). Freebase: a collaboratively created
graph database for structuring human knowledge. In Proceedings of the 2008 ACM SIGMOD Interna-
tional Conference on Management of Data (pp. 1247−1250). Association for Computing Machinery.
Gormley, M.R., Yu, M. & Dredze, M. (2015). Improved relation extraction with feature-rich composi-
tional embedding models. arXiv preprint arXiv:1505.02419.
Gupta, R. & Sarawagi, S. (2011). Joint training for open-domain extraction on the web: exploiting over-
lap when supervision is limited. In Proceedings of the fourth ACM International Conference on Web
Search and Data Mining (pp. 217−226). Association for Computing Machinery.
Hochreiter, S. & Schmidhuber, J. (1997). Long short-term memory. Neural Computation, 9(8),
1735−1780.
Hoffmann, R., Zhang, C., Ling, X., Zettlemoyer, L. & Weld, D.S. (2011). Knowledge-based weak super-
vision for information extraction of overlapping relations. In Proceedings of the 49thAnnual Meeting
of the Association for Computational Linguistics: Human Language Technologies-Vol 1 (pp. 541−550).
Lample, G., Ballesteros, M., Subramanian, S., Kawakami, K. & Dyer, C. (2016). Neural architectures
for named entity recognition. arXiv preprint arXiv:1603.01360.
Li, Q. & Ji, H. (2014). Incremental joint extraction of entity mentions and relations. In ACL (1)
(pp. 402−412).
Min, B., Grishman, R., Wan, L., Wang, C. & Gondek, D. (2013). Distant supervision for relation extrac-
tion with an incomplete knowledge base. In HLT-NAACL (pp. 777−782).
Mintz, M., Bills, S., Snow, R. & Jurafsky, D. (2009). Distant supervision for relation extraction with-
out labeled data. In Proceedings of the Joint Conference of the 47th Annual Meeting of the ACL
and the 4th International Joint Conference on Natural Language Processing of the AFNLP, Vol 2
(pp. 1003−1011). Association for Computational Linguistics.
Ren, X., Wu, Z., He, W., Qu, M., Voss, C.R., Ji, H., ... & Han, J. (2017, April). CoType: Joint extraction
of typed entities and relations with knowledge bases. In Proceedings of the 26th International Con-
ference on World Wide Web (pp. 1015−1024). International World Wide Web Conferences Steering
Committee.
Riedel, S., Yao, L. & McCallum, A. (2010). Modeling relations and their mentions without labeled text.
Machine Learning and Knowledge Discovery in Databases, 148−163.
Ritter, A., Zettlemoyer, L. & Etzioni, O. (2013). Modeling missing data in distant supervision for infor-
mation extraction. Transactions of the Association for Computational Linguistics, 1, 367−378.
Surdeanu, M., Tibshirani, J., Nallapati, R. & Manning, C.D. (2012). Multi-instance multilabel learning
for relation extraction. In Proceedings of the 2012 Joint Conference on Empirical Methods in Natural
Language Processing and Computational Natural Language Learning (pp. 455−465). Association for
Takamatsu, S., Sato, I. & Nakagawa, H. (2012). Reducing wrong labels in distant supervision for rela-
tion extraction. In Proceedings of the 50th Annual Meeting of the Association for Computational
Linguistics: Long Papers-Vol 1 (pp. 721−729). Association for Computational Linguistics.
Wang, C., Fan, J., Kalyanpur, A. & Gondek, D. (2011, ). Relation extraction with relation topics. In Pro-
ceedings of the Conference on Empirical Methods in Natural Language Processing (pp. 1426−1436).
Xu, W., Hoffmann, R., Zhao, L. & Grishman, R. (2013, ). Filling knowledge base gaps for distant super-
vision of relation extraction. In ACL (2) (pp. 665−670).
Zheng, S., Wang, F., Bao, H., Hao, Y., Zhou, P. & Xu, B. (2017). Joint extraction of entities and relations
based on a novel tagging scheme. arXiv preprint arXiv:1706.05075.
Zheng, S., Xu, J., Zhou, P., Bao, H., Qi, Z. & Xu, B. (2016). A neural network framework correlation
extraction: Learning entity semantic and relation pattern. Knowledge-Based Systems, 114, 12−23.
850

Decoupling control of TRMS based on a Relative Gain Array

(RGA) and Kharitonov theorem
Sumit Kumar Pandey, Jayati Dey & Subrata Banerjee

Department of Electrical Engineering, National Institute of Technology, Durgapur, India
ABSTRACT: This work addresses decoupling control of TRMS. The modeling of TRMS
is done through identification algorithm using real-time input-output data. A decoupling
technique is proposed on basis of Relative Gain Array (RGA) to eliminate the cross coupling
effect. The stabilization of TRMS is achieved with PID controllers, the parameter range
of which is obtained with Kharitonov stability criteria. PSO method is further tested for
parameters tuning within range obtained earlier. The performance of the controller is tested
in simulation and responses are found satisfactory.
1 INTRODUCTION
The TRMS has two control inputs with two outputs, respectively, as pitch (ψ) and yaw (ϕ)
with significant cross coupling between them [1]. The designing of the controller is always
a challenging task due to strong coupling effect and highly nonlinear characteristics of the
TRMS.
In [2] authors explain method of identification for TRMS. The non-linear least squares
identification method is applied for calibration. In [3] identification is performed by neural
network approaches for TRMS. The different methods of identification and control tech-
nique which are frequently applied in MIMO system is explained in [5, 6]. The relative gain
analysis is used for the pairing analysis of the MIMO system [7] and the decoupling tech-
nique is applied to eliminate the cross coupling effect of MIMO system.
The aim of the present work is to design PID controller to control the decoupled TRMS
plant. First the identification of the TRMS is done and the transfer function is obtained.
The decoupling of the TRMS is performed to eliminate the cross coupling effect of the
TRMS and this decoupling is validated by the RGA analysis and the simulation results. The
PID controller is designed on the basis of Kharitonov theorem through which a robust PID
controller range is obtained. Further PSO is employed for the fine tuning of the controller
parameters. The remainder is as next section describes the identification process of TRMS
followed by the decoupling technique in section 3. In section 4 range of PID parameters are
determined using Kharitonov theorem while in section 5 PSO method is implemented to
obtain the optimum value of PID parameter. Section 6 depicts the simulation results and at
last conclusion.
2 IDENTIFICATION OF TRMS MODEL
The objective of identification process to find the transfer function of TRMS. As there exist
cross-coupling in two rotors of TRMS, it is considered as two linear rotor models with two
linear couplings in-between. Therefore four linear models have to be identified as u1 to y1 and
u2 to y2 u2 to y1 and u1 to y2 as shown in Figure 1. An experiment for model identification is
carried out with the help of the MATLAB Toolbox using chosen identification models for
four transfer functions [4–6]. The TRMS model and the experimental setup are excited with
851

Figure 1. Transfer function model for TRMS.
same input excitation and their responses are recorded. The excitation signal contains differ-
ent sinusoids. To estimate accurate model error is minimized between the chosen model and
the actual plant output. The optimal model parameters, for which the square of the error is
minimal (“least mean square (LMS)” method) is considered as “identified model” [7]. The
identified model of TRMS is obtained as,
 G11 ( s ) G12 ( s ) 
G (s) =   (1)
 G21 ( s ) G22 ( s ) 
 0.01657 s 2 + 0.4194 s + 2.454 0.04986 s . 
 s 3 + 1.487 s 2 + 4.403s + 5.449 2
. s + 0 0962
. 
G (s) =   (2)
 . . . s + s02 2377
+ s + 4 s902+ 
 20 02248s + 0 4527
. . 0s.0009881
3
+ s 2 + 0.03361s + 0.4065 
s + 0 4099s + 0 2181 1.345 0.4568 0.3826
3 DECOUPLING OF TRMS
Here a decoupler is designed for TRMS based on generalized decoupling technique to rectify
the coupling effect associated with the plant. In the generalized decoupling technique the design
of decoupler for any square plant G(s) is based on the formula as described by equation 3. The
RGA method is applied to investigate the pairing analysis of the plant and it is verified that the
decoupled plant GN(s) shown by Figure 2 is perfectly decoupled as described by equation (7).
GD ( s ) = GI (0 ) * GR ( s ) (3)
where, G(0) = steady state gain matrix of G(s), GD(s) = Decoupling matrix of G(s),
GI(0) = inverse matrix of G(0) & GR(s) = Diagonal matrix of G(s). Considering G11,G22 only,
the diagonal form of the plant.
 0.01657 s 2 + 0.4194 s + 2.454 

 s 3 + 1.487 s 2 + 4.403s + 5.449  0
GR ( s ) =   (4)
 0.000 s + .
2
s+ . 
 0
s 3 +09881
. s + .0 4065 
s 2 +0 03361
.
1 395 0 456 0 3826
The decoupler is designed by following equation (3) as,
 0.04 s 2 + 1.01s + 5.93 −0. s2 − . s− . 

 s 3 + 1.487 s 2 + 4.403s + 5.449 0 000039524
s + .
3
s + . 0 01626 
s 2 + 0. 00134
D ( s ) = G I ( 0 ) * GR ( s ) =  
 − .0 s − . s − .
2
. 1 395 s 2 0+ 456
. 0 0s +3826
. 
 s 3 +0 07
. s 2 1+ 97. 11
s + 582
. 0s001007862
3
+ . s + .
2 0 0034
s + . 41
0  (5)
1 487 4 403 5 449 1 395 0 456 0 3826
852

Figure 2. Decoupled plant of TRMS.
Figure 3. Output response of decoupled plant when step input is applied to yaw rotor while zero input
is applied to pitch rotor.
Figure 4. Output response of decoupled plant when step input is applied to pitch rotor while zero
input is applied to yaw rotor.
The decoupled plant is determined as below.

GN ( s ) = G ( s ) * D ( s ) (6)
Now, the relative gain array of the above matrix is calculated as described by equation (7).
It verifies that upon decoupling the each output of the plant solely depends on the individual
single reference input.
  0.4457 0.0018    1 0 
RGA(GN (0 )) = RGA(G (0 ) * D(0 )) =    =  (7)
  −0.0116 1.0196    0 1 
It is also verified by simulation results as shown in Fig. 3 when step is tested to port 2 of
the decoupler (Fig. 2) and zero input is tested to port 1. The corresponding yaw output shows
step response as shown in Fig. 3 while pitch shows zero response signifying that decoupler
completely nullifies the coupling effect on output. Similar situation arises when a step is
tested to port 1 of the decoupler and zero input to port 2 as result of which corresponding
pitch output shows step response while yaw shows zero response shown in Fig. 4.
4 APPLICATION OF KHARITONOV THEOREM TO FIND THE RANGE OF

PARAMETERS OF PID CONTROLLER
Let assume the set δ(s) of real polynomials of degree n of the form as
δ ( s ) = δ 0 + δ1s + δ 2 s 2 + δ 3s 3 + δ 4 s 4 + .........δ n s n (8)
853

The coefficient is in between as δ 0 ∈[ x0 , y0 ],δ1 ∈[ x1, y1 ],......δ n ∈[ xn , yn ]
Set δ(s) is said to be Hurwitz if following polynomial is Hurwitz.
K1 ( s ) = x0 + x1s + y2 s 2 + y3s 3 + x4 s 4 + x5 s 5 + y6 s 6 + .......,

K 2 ( s ) = x0 + y1s + y2 s 2 + x3s 3 + x4 s 4 + y5 s 5 + y6 s 6 + .......,
K3 ( s ) = y0 + x1s + x2 s 2 + y3s 3 + y4 s 4 + x5 s 5 + x6 s 6 + ......., (9)
K 4 ( s ) = y0 + y1s + x2 s 2 + x3s 3 + y4 s 4 + y5 s 5 + x6 s 6 + .......,
PID controller transfer function is written as below,
Ki
H (s) = K p + + Kd s (10)
s
On the basis of above equation the characteristic equation 1 + G11 ( s )H ( s ) is written as

below.
2.454 K i + s(2.454 K p + 0.4194 K i + 5.449)

+ s 2 (0.4194 K p + 0.01657 K i + 2.454 K d + 4.40 ) (11)
+ s 3 (1.487 + 0.01657 K p + 0.4194 K d ) + s 4 (1 + 0.01657 K d )
Now the four interval polynomial has been obtained from the above equation where the
range of K p , K i & K d are ( K p− K p+ ),( K i− K i+ )& ( K d− K d+ ) respectively.
K1 ( s ) = 2.454 K i− + s(2.454 K p− + 0.4194 K i− + 5.449)

+ s 2 (0.4194 K p+ + 0.1657 K i+ + 2.454 K d+ + 4.40 )
+ s 3 (1.487 + 0.01657 K p+ + 0.4194 K d+ ) + s 4 (11 + 0.01657 K d− )
K 2 ( s ) = 2.454 K i+ + s(2.454 K p+ + 0.4194 K i+ + 5.449)

+ s 2 (0.4194 K p− + 0.1657 K i− + 2.454 K d− + 4.40 )
+ s 3 (1.487 + 0.01657 K p− + 0.4194 K d− ) + s 4 (11 + 0.01657 K d+ )
(12)
K3 ( s ) = 2.454 K i+ + s(2.454 K p− + 0.4194 K i− + 5.449)
+ s 2 ( 0.4194 K p− + 0.1657 K i− + 2.454 K d− + 4.40 )
+ s 3 (1.487 + 0.01657 K p+ + 0.4194 K d+ ) + s 4 (11 + 0.01657 K d+ )
K 4 ( s ) = 2.454 K i− + s( 2.454 K p+ + 0.4194 K i+ + 5.449)

+ s 2 (0.4194 K p+ + 0.1657 K i+ + 2.454 K d+ + 4.40 )
+ s 3 (1.487 + 0.01657 K p− + 0.4194 K d− ) + s 4 (11 + 0.01657 K d− )
In order to satisfy the condition of stability for the above four intervals polynomial the
range of PID controller parameter is obtained as in case of the main rotor are Kp = [0.1−1],
Ki = [0.1−1] & Kd = [0.5−2]. Similarly, by adopting the same procedure the range of PID
controller paramet. obtained for tail rotor is as Kp = [0.1−2], Ki = [0.1−0.5] & Kd = [2−5].
5 APPLICATION OF PSO TECHNIQUE TO TUNE THE PID CONTROLLER

PARAMETERS
The PSO algorithm is described by flow chart in Figure 5. The velocity and position formula
is calculated here is as below.
Vidn+ 1 = Vidn + c1rand ().( Pidn − X idn ) + c2rand ().( Pgdn − X idn ) (13)
854

Figure 5. PSO algorithm flow chart.
Figure 6. Convergence characteristics of main

rotor. Figure 7. Convergence characteristics of tail
rotor.
Figure 8a. Step response of the main rotor.

Figure 8b. Control signal of main rotor.
X idn+ 1 = X idn + Vidn+ 1 (14)
The objective function in the time domain is calculated as W [8]
W = (1 − e− β ).(M p + E ss ) + e− β (ts − tr ) (15)
The convergence characteristic of main and tail rotor is shown in Figures 6 and 7 respectively.
855

Figure 9b. Control signal of tail rotor.
Figure 9a. Step response of the tail rotor.
6 SIMULATION RESULTS
The parameter value of the PID controller is individually determined for both rotors using
PSO technique within the ranges as obtained by Kharitonov criteria. The step response and
the control signal are exhibited by Figure 8(a) and 8(b) and 9(a) and 9(b), respectively, for
main and tail rotors.
7 CONCLUSION
This paper designs a decoupler for TRMS based on RGA technique in order to nullify the
cross-coupling effect of the MIMO system. Then two PID controllers derived based on the
Kharitonov stability theorem have been tuned using PSO technique. The accuracy of this
method is established by results. To further validate the simulation results, the proposed tech-
nique will be implemented in the real time system of TRMS at NIT Durgapur Advanced
Control laboratory.
REFERENCES
[1] TRMS 33–949S User Manual. Feedback Instruments Ltd., East Sussex, U.K., 2006.
[2] D. Rotondo, F. Nejjari, V. Puig (2013) Quasi-LPV modeling, identification and control of a twin
rotor MIMO system, Control Engineering Practice, vol. 21, iss. 6, pp. 829–846.
[3] B. Subudhi & D. Jena (2009) Nonlinear system identification of a twin rotor MIMO system, TEN-
CON 2009 IEEE Region 10 Conference.
[4] M. A. Hossain, A. A. M. Madkour, K. P. Dahal & H. Yu, (2004) Intelligent active vibration control
for a flexible beam system, Proceedings of the IEEE SMC UK-RI Chapter Conference, London-
derry, U.K.
[5] I. Z. Mat Darus & Z. A. Lokaman (2010) Dynamic modeling of Twin Rotor Multi System in hori-
zontal motion, Journal Mekanikal, no. 31, pp. 17–29.
[6] I. Z. Mat Darus (2004) Soft computing active adaptive vibration control of flexible structures, Ph.D.
Thesis, Department of Automatic Control and System Engineering, University of Sheffield.
[7] A. Rahideh, M.H. Saheed & H.J.C. Huijberts (2008) Dynamic modeling of the TRMS using Ana-
lytical and Empirical approaches”, Control Engineering Practice, vol. 16, no. 3, pp. 241–259.
[8] Z. L.Gaing (2004) A Particle Swarm Optimization Approach for optimum Design of PID Control-
ler in AVR System, IEEE transcations on Energy Conversion, vol. 19, no. 2, pp. 384–391.
856

Root ORB—an improved algorithm for face recognition
A. Vinay, Aprameya Bharadwaj, Arvind Srinivasan, K.N. Balasubramanya Murthy

& S. Natarajan
Center for Pattern Recognition and Machine Intelligence, PES University, Bangalore, India
ABSTRACT: The biggest challenge faced by Face Recognition is to extract important

facial features efficiently. Existing methods like SIFT, Root SIFT perform quite accurately,
but fail in real world applications that require real-time processing. The ORB feature detec-
tor-descriptor performs quite well in this regard. We have modified the ORB feature detector
to Root ORB and used it as a pre-processing step for a study of classification algorithms for
face recognition. Our results show that the Root SIFT feature detector-descriptor performs
only 6.25% to 12.5% as fast as the Root ORB detector-descriptor. We have chosen to work
with detector-descriptors like ORB rather than using Neural Networks to extract facial fea-
tures. Compute-intensive methods like Neural Networks require a lot of training data and
computational time to obtain these features.
1 INTRODUCTION
Face recognition is the method or technique of detecting and recognizing a person from an
image. A popular approach is to extract features from an image and use these to match with
other images.
It has many applications in security, identification systems, and surveillance. For example,
the FBI has a program to include face recognition along with other biometrics to retrieve
records from its database. We also find everyday uses of face recognition like authentication
mechanisms in mobile devices.
Face recognition can be traced back to the 1960s. The objective then, was to select a small
set of images from a large database that could possibly contain an image to be matched.
Since then, we’ve come a long way. Now, features extracted from a single image can give us a
good idea about the image. We’ve been able to overcome hurdles like change in illumination,
change in facial expressions, and motion of the head. Also, it is relatively easier to obtain
datasets today.
We have modified the existing ORB detector-descriptor to Root ORB in the hope of
improving accuracy and decreasing computational time. We have used the Bag of Words
model before classification to create image histograms. We have then used machine learning
classifiers to classify the images in the dataset. We have conducted a comparative study of
Root SIFT versus Root ORB. ORB was used as it is very fast and open source.
2 TECHNICAL BACKGROUND
In face recognition, faces in the image are identified and represented by features and descrip-
tors extracted from the image. This involves two steps: Feature detection and Description.
In detection, points of interest are determined. In description, attributes of these points
are ascertained and stored in a vector. This vector is then used for applications like image
classification. The computational efficiency of this task depends on the feature detector-
descriptor algorithms we use, and the learning algorithms we apply.
857

Popular detector-descriptor methods include: Scale Invariant Feature Transform (SIFT)
(Lowe (2004)), Speeded-Up Robust Features (SURF) (Bay et al 2008)), Oriented FAST and
Rotated BRIEF (ORB) (Rublee et al. (2011)). Another important feature detector-descriptor
is a modification of SIFT called Root SIFT (Arandjelović et al. (2012)). The main difference
between these are the size of the feature vector obtained and the computational time.
SIFT returns a 128-dimensional vector of features. SURF takes lesser time as compared to
SIFT as it returns a 64-dimensional vector of features. ORB is the fastest as it uses Features
from Accelerated Segment Test (FAST) (Rosten et al. (2006)) detector and Binary Robust
Independent Elementary Features (BRIEF) (Calonder et al. (2010)) descriptor and returns a
32-dimensional vector.
Root SIFT improves the accuracy of SIFT but still requires a similar computational time
as it also returns a 128-dimensional vector.
Here, we use Root ORB detector-descriptor and some machine learning classifiers on the
images of two well-known datasets. As Root ORB returns a 32-dimensional feature vector,
we hope to achieve accuracy in the neighbourhood of Root SIFT when used along with these
classifiers and optimize on the computational time.
The machine learning classification algorithms we use are:
a. Support Vector Machine Algorithm using Additive Chi Squared Kernel
b. K-Nearest-Neighbors Algorithm
c. Naïve Bayes Algorithm.
2.1 SIFT
The scale-invariant feature transform (SIFT) is an algorithm to extract and characterize local
features in images. SIFT is scale and rotation invariant, and is therefore capable of delivering
promising accuracies. It extracts the key-points within the image. These key points are then
used for comparison, classification and matching. It performs the following functions to
extract the key points. First, it identifies the interest points on the image by using Gaussian
Difference. Then, the location and scale for these interest points are determined. Next, an ori-
entation is assigned to each of these interest points. Finally, gradients are measured around
these points and the image is transformed to minimize distortion.
2.2 Root SIFT

For certain areas in computer vision, using Euclidean distance to measure similarity between
descriptors often yields inferior performance. In these areas, using the Hellinger kernel yields
a superior performance when compared to using Euclidean distances. Root SIFT is a descrip-
tor which is an element wise square root of the L1 normalized SIFT vectors.
2.3 ORB
The ORB algorithm makes use of the FAST detector and BRIEF descriptor. First, FAST is
applied to obtain all key points. Then, the Harris corner measure is applied and the best key
points among them are found. ORB makes use of the BRIEF descriptor. Although BRIEF
performs poorly on rotation, ORB stabilizes it by using the generated key points.
2.4 FAST
FAST detector uses circles to classify key points as corner points. To check whether a pixel
is a corner point, we compare the intensity of the pixel with the intensities of the circle of
points around it. If they are considerably brighter or darker, we can flag this pixel as a corner
point. FAST is renowned for its high computational efficiency. It also performs efficiently
when used with machine learning algorithms.
858

2.5 BRIEF
SIFT uses 128-dimensional float vector for descriptors. Similarly, SURF uses 64-dimensional
float vector. Working with these vectors for a large number of features requires a lot of space.
It also requires a long time to match these features. Furthermore, all of these dimensions are
not required for matching. BRIEF is a feature descriptor which is used along with feature
detectors like FAST. It is generally used in applications which require real time processing.
3 PROPOSED METHOD
In subsequent paragraphs, we discuss in detail the methods used in our proposed system.
3.1 Root ORB

We applied the existing root algorithm from Root SIFT to the ORB algorithm. Here, as men-
tioned earlier, using the Hellinger kernel yields a superior performance when compared to
using Euclidean distances. Hellinger kernel for L1 normalized histograms x and y is given by:
n
H ( x, y) = ∑ xi yi (1)
i =1
After L1 normalization is done to all the ORB vectors, element wise square root is taken
for all these vectors.
In the feature map space, calculating the Euclidean distance is analogous to calculating the
Hellinger distance in the original space as:
x ′T y ′ = H ( x , y ) (2)
Where x’ is the ORB vector after L1 normalization and computing element wise square
root.
Key points are obtained for the images after Root ORB is applied. Root ORB uses a 32-
dimensional vector to store the descriptors obtained and also uses the FAST detector. Due to
this, the time required for classification is much lesser than compared to Root SIFT.
3.2 Bag of words model

In face recognition, we can represent a face by a bag of words. These visual words are a
discriminative representation of the key points extracted from the face. These key points are
Figure 1. When Root ORB is applied to images of the datasets.
859

encoded by a visual vocabulary. Next, all these features are pooled into an image represen-
tation. We use this model by building a large vocabulary of visual words. We analyze each
image by observing the frequency of the words appearing in the image.
3.3 Support vector machine

Support vector machines (SVM) (Hearst et al. (1997)) are supervised learning models which
are generally used for classification. When you provide data that consists of data points
belonging to different, known categories, the SVM algorithm builds a model which can
assign categories to new data points. When you represent the SVM model in space, you see
that the algorithm works in such a manner as to maximise the gap between the different
classes. A new data point added to this space can be classified by determining the category
which is closest to this data point.
SVMs, by default, use linear classification methods. Using different kernels, we can also
perform non-linear classification.
The additive chi squared kernel (Maji et al. (2008)) is often used in computer vision. Since
the kernel is additive, it is possible to treat all components separately. This makes it possible
to sample the Fourier transform in regular intervals. It is given by:
2 xi yi
k ( x, y ) = ∑ (3)
i xi + yi
3.4 Naïve bayes

Naive Bayes (Lewis (1998)) is a set of algorithms used for classifying features which uses the
Bayes’ theorem of probability. The features being classified are assumed to be independent
of each other.
Naïve Bayes is scalable and can be extended to a large number of features.
Here we make use of the Gaussian Naïve Bayes where the likelihood of the features is
assumed to be Gaussian in nature. It is given by:
 ( ) 
2
1 xi − µ y
P ( xi | y ) = exp  −  (4)
2πσ y
2
 2σ y2 
The parameters σy and µy are estimated using maximum likelihood.
3.5 K-nearest neighbors

In machine learning, K-Nearest Neighbors (Guo et al. (2003)) algorithm is used for clas-
sification. The output indicates whether it belongs to a certain class or not. For a given data
point, its K-nearest neighbors are determined. Then, the class that occurred most frequently
amongst these neighbors is assigned to the data point. If k = 1, then the object is assigned to
that particular class. If k = 0 then it indicates that it doesn’t belong to that class.
3.6 Pipeline
After Root ORB was applied to the dataset, the descriptors obtained were clustered using
the bag of words model. Then, the model was separated into a training set and a testing set.
Finally, we studied the results obtained from each of the three classifiers we used – K-NN,
Naïve Bayes, SVM. These results were then tabulated. This was done for the Faces95 and
Grimace datasets.
860

Figure 2. An overview of the entire process.
Figure 3. Sample of Faces95 dataset [11].
Figure 4. Sample of Grimace dataset [12].
4 DATASETS
The datasets used are:

1. Faces95 dataset – This dataset contains of 72 different people with 20 pictures each (both
male and female). The size of each image is 180 × 200. The background is red. There is
large head scale variation and change in lighting between images. There is slight variation
of expression and motion of face in the images.
2. Grimace dataset – This dataset contains of 18 different people with 20 pictures each (both
male and female). The size of each image is 180 × 200. The background is plain. There is
little head scale variation and change in lighting between images. There is a large variation
of expression and motion of face in the images.
5 RESULTS
All our programs were executed on Intel i7 core, using Windows10. The packages which were
used are Opencv3 (3.2.0), scikit-learn (0.18.1), numpy (1.11.0), joblib (0.11) and inbuilt pack-
ages in python which include OS, time, random and math. Python3.5 was used.
First, we present the results for Root SIFT when applied with K-NN, Naïve Bayes, and
SVM classification algorithm:
1. When Root SIFT is used on Grimace database with the following machine learning algo-
rithms applied independently:
2. When Root SIFT is used on Faces95 database with the following machine learning algo-
3. When Root ORB is used on Grimace database with the following machine learning algo-
4. When Root ORB is used on Faces95 database with the following machine learning algo-
861

Table 1. Root SIFT for Grimace dataset.
Method Precision Recall F1-score Accuracy Time taken
K-NN 0.92 0.89 0.88 0.88 16.11

Naïve Bayes 0.92 0.87 0.86 0.86 15.80
SVM 0.98 0.97 0.97 0.97 15.78
Inference: The SVM Classifier was the most accurate.
Table 2. Root SIFT for Faces95 dataset.
K-NN 0.94 0.91 0.91 0.90 140.08

Naïve Bayes 0.83 0.76 0.76 0.76 53.93
SVM 0.99 0.99 0.99 0.98 61.09

The K-NN classifier was the slowest.
Table 3. Root ORB for Grimace dataset.
K-NN 0.89 0.87 0.85 0.86 1.52

Naïve Bayes 0.88 0.87 0.86 0.86 1.49
SVM 0.99 0.99 0.99 0.98 1.59
Table 4. Root ORB for Faces95 dataset.
K-NN 0.89 0.87 0.86 0.87 8.69

Naïve Bayes 0.94 0.91 0.91 0.90 7.71
SVM 0.97 0.97 0.97 0.97 8.12
6 CONCLUSIONS
The Faces95 dataset poses the challenges of large head scale variation and illumination
changes, and the Grimace dataset focusses on large variation in expression and the motion
of the face. From our results, we see that the accuracy of our method is high in all our test
cases. Hence, we can say that our method overcomes the above hurdles.
From Table 2, we notice that the K-NN Classifier has performed much slower as com-
pared to the other classifiers. Upon closer observation, we see that it performs the slowest
in all the cases. This could be attribute to the fact that finding the nearest neighbor in a high
dimensionality space requires a lot of time.
From the results, we notice that the SVM Classifier is the most accurate classifier. This
could be due to the Additive Chi Squared Kernel which applies Fourier Transforms at regular
intervals to the dataset.
When we compare Tables 1 and 2 with Tables 3 and 4 respectively, we observe that the
accuracy of the Root ORB method is about the same, or sometimes greater than the accuracy
of the Root SIFT method. However, when the computational times are compared, the Root
ORB method is significantly faster than the Root SIFT method.
862

Also, the key point vector size in the Root ORB method is 75% smaller than the vector size
in the Root SIFT method.
Hence, we see that the Root ORB method is both space and time efficient, in obtaining key
points for classifying images as compared to the Root SIFT method.
REFERENCES
Arandjelović, Relja, and Andrew Zisserman. “Three things everyone should know to improve object
retrieval.” In Computer Vision and Pattern Recognition (CVPR), 2012 IEEE Conference on, pp. 2911–
2918. IEEE, 2012.
Bay, Herbert, Tinne Tuytelaars, and Luc Van Gool. “Surf: Speeded up robust features.” Computer
vision–ECCV 2006(2006): 404–417.
Calonder, Michael, Vincent Lepetit, Christoph Strecha, and Pascal Fua. “Brief: Binary robust inde-
pendent elementary features.” Computer Vision–ECCV 2010 (2010): 778–792.
Faces95 Dataset—http://cswww.essex.ac.uk/mv/allfaces/faces95.html.
Grimace dataset—http://cswww.essex.ac.uk/mv/allfaces/grimace.html.
Guo, Gongde, Hui Wang, David Bell, Yaxin Bi, and Kieran Greer. “KNN model-based approach in
classification.” In CoopIS/DOA/ODBASE, vol. 2003, pp. 986–996. 2003.
Hearst, Marti A., Susan T. Dumais, Edgar Osuna, John Platt, and Bernhard Scholkopf. “Support vec-
tor machines.” IEEE Intelligent Systems and their applications 13, no. 4 (1998): 18–28.
Lewis, David D. “Naive (Bayes) at forty: The independence assumption in information retrieval.”
In European conference on machine learning, pp. 4–15. Springer, Berlin, Heidelberg, 1998.
Lowe, David G. “Distinctive image features from scale-invariant keypoints.” International journal of
computer vision60, no. 2 (2004): 91–110.
Maji, Subhransu, Alexander C. Berg, and Jitendra Malik. “Classification using intersection kernel sup-
port vector machines is efficient.” In Computer Vision and Pattern Recognition, 2008. CVPR 2008.
IEEE Conference on, pp. 1–8. IEEE, 2008.
Rosten, Edward, and Tom Drummond. “Machine learning for high-speed corner detection.” Computer
Vision–ECCV 2006(2006): 430–443.
Rublee, Ethan, Vincent Rabaud, Kurt Konolige, and Gary Bradski. “ORB: An efficient alterna-
tive to SIFT or SURF.” In Computer Vision (ICCV), 2011 IEEE international conference on,
pp. 2564–2571. IEEE, 2011.
863

Face recognition using SURF and delaunay triangulation
A. Vinay, Abhijay Gupta, Harsh Garg, Shreyas Bhat, K.N. Balasubramanya Murthy &
S. Natarajan
Center for Pattern Recognition and Machine Intelligence, PES University, Bangalore, India
ABSTRACT: Face recognition is transforming the way people are interacting with
machines. Earlier it was used in specific domains like law enforcement but with extensive
research being done in this field, it is being extended to various applications like automatic
face tagging in social media, surveillance systems in airports, theaters and so on. Local fea-
ture detection and description is gaining a lot of significance in the face recognition commu-
nity. Extensive research on SURF and SIFT descriptors have found widespread application.
Key points matched by SURF and triangulated using Delaunay Triangulation boosts the
interest points detected. Other modern techniques like machine learning and deep learn-
ing require huge amount of training data and computational capabilities which sometime
becomes a limitation of its usage. In contrast, hand-crafted models like SURF, SIFT over-
comes the requirement of training data and computing power. The pipeline proposed in the
paper reduces the average computational power and increases accuracy.
1 INTRODUCTION
One of the most challenging problem faced by the computer vision community is Face Rec-
ognition. Decades of work in this field has made face recognition system almost as smart
as human beings. Face Recognition has been increasingly applied in surveillance systems
and authentication services in order to prevent loss of sensitive information and curb secu-
rity breaches leading to loss of money. Apart from these applications, Face Recognition is
also extensively being used by social networking sites such as Facebook to tag friends. Tech
giants such as Google, Microsoft use face recognition for image based search and to provide
authentication services respectively.
Although face recognition has been effectively used in a number of applications, its per-
formance tend to decline when the image shows significant variations in pose, expression,
scale, illumination and translation. Most of the real world problems that require face recogni-
tion possesses these variations. Computer Vision enthusiasts are trying to make this system
invariant to all these challenges. To overcome the problems associated with pose, expres-
sion, scale and upto some extent illumination, we propose a robust model which performs
well when experimented with datasets which contain images variant to above mentioned
constraints.
In any face recognition system, the most crucial step is to locate interest points in an image.
A vast variety of keypoint detectors and feature descriptors have been extensively used by
researchers in literature (eg. Bay et al. (2008), Lindeberg (1998), Lowe (2004)). In recent
years, considerable amount of work has been done on Speeded-Up Robust Features (SURF)
which is built upon previous works (eg. SIFT) to speed up and incorporate invariance in scale
and in-plane rotation. The proposed model combines several algorithms and mathematical
functions to boost the robustness and veracity of the system.
865
CH147_040.indd 865 7/9/2018 4:29:03 PM

2 RELATED WORK
Delaunay Triangulation find its use in finger print identification, logo recognition and other
object recognition applications (e.g Miri and Shiri (2012)). In Bebis et al. (1999) the Delaunay
Triangulation is computed using ridge endings and ridge bifurcation minutiae represented
in form of its coordinates. This index-based new approach achieves average accuracies of
86.56%, 93.16%, 94.12% for 3, 5, 7 imprints per person on testing set of size 210, 150 and 90
respectively. It characterizes better index selectivity, low storage requirements, minimal index-
ing requirements and fast identification for fingerprint recognition. Kalantidis et al. (2011)
uses a novel discriminative triangle representation using multi-scale Delaunay Triangulation.
Indexing is done using inverted file structure for robust logo recognition. On adding 4k dis-
tractor classes to the Flickr dataset the performance of the proposed multi-scale Delaunay
Triangulation approach drops by 5.5% as compared to the base line Bag of Words model.
SIFT (Geng and Jiang (2009a), Geng and Jiang (2009b)) introduced enhancements like
Partial-Descriptor-SIFT, Key-Points-Preserving-SIFT and Volume-SIFT (VSIFT) by keep-
ing all the initial keypoints, preserving the interest points on a large scale or near face bound-
aries and by removing unreliable keypoints based on their volume respectively. They reduce
the error rate by 4.6% and 3.1% in KPSIFT and PDSIFT on the AR dataset. SIFT in com-
bination with a bag of words model has been deployed in (Sampath et al. (2016)) to detect
household objects. Modern techniques such as Convolutional Neural Network is outper-
formed by a hybrid model (Al-Shabi et al. (2016))) using SIFT.
The effectiveness of combining Delaunay Triangulation and SIFT has been demonstrated
in Dou and Li (2012). With the increase in number of viewpoints, the accuracy of SIFT algo-
rithm decreases. To overcome this, Delaunay Triangulation (DelTri) exploits the overlapped
regions in different images. Due to uniqueness of the DelTri, the overlapped region of an
image pair overcomes the changes which were generated by different viewpoints at certain
degree, thus increases the correct ratio of the result matches from 58.8% (SIFT+RANSAC)
to 88.2% (SIFT+DelTri). This combination has also been used in Liu et al. (2017) for face
alignment along with Convolutional Neural Networks.
Recent descriptors such as SURF is improvised upon SIFT by restricting the total number
of reproducible orientation through the use of information from circular area around the
interest points. After this step, construction of square region is performed. Further, the
descriptors are extracted from these regions. SURF is used in various object tracking algo-
rithms (Shuo et al. (2012)), where interest points in the defined object are matched between
consecutive frames by obtaining the Euclidean distance between their descriptor.
3 METHOD PROPOSED
A robust approach is proposed with better results by combining Bilateral Filters for image
smoothing, SURF to detect facial keypoints, PCA to minimize the number of key points,
FLANN and Delaunay Triangulation for face matching in the dataset. The main steps of
our model is depicted in Fig. 1.
3.1 Bilateral filter

Filtering is mostly used for reducing noise in the nearby pixel values which are mutually less
correlated than the signal values. Tomasi and Manduchi (1998) smoothen the image by utilizing
Figure 1. Proposed approach – BF + SURF + PCA + FLANN.
866

a non-linear combination derived by averaging the smooth regions of the nearby image values,
preserving edges. Other filtering techniques such as Gaussian blurring (Mostaghim et al.
(2014)) which performs linear operation, neglecting edges is determined by sigma and non-
linear filters like median filter (Dreuw et al. (2009)) which replaces the pixel values with the
median value available in the local neighborhood are outperformed by Bilateral filters. In
Bilateral Filtering weight of the intensity values of the surrounding pixels replace the inten-
sity value of a pixel that is based on Gaussian distribution. This system loops over each pixel
and simultaneously adjust weights of the neighboring pixels, retaining sharp edges.
Bilateral filtering combines the range and spatial domain filters
∞ ∞
1
h( x ) = ∫ −∞∫ g(ξ )c(ξ , x )s( g(ξ ), g( x ))d
n( x ) −∞
(1)
where is n(x) the normalization factor, calculated as

∞ ∞
1
n( x ) = ∫ −∞∫ c(ξ , x )s( g(ξ ), g(x ))dξ
n( x ) −∞
(2)
In which c(ξ, x) evaluates the geometric closeness between the neighboring center x and
a nearby point ξ. s(g(ξ), g(x)). evaluates the photometric similarity between the pixel at the
neighborhood center x and that of a nearby point ξ.
3.2 SURF
Speeded-up robust features is invariant to in-plane rotation, contrast, scale and brightness.
It comprises of keypoint point detector which interpolates the highly discriminative facial
points. Further the descriptor extracts the features of the keypoint by constructing feature
vectors. To minimize computation time SURF employs fast Hessian Matrix approximation
and the scale space is examined by up scaling the integral image based filter sizes to detect
interest points.
Given a point a = (x, y) in an image I, the Hessian matrix H(a; σ) in at scale is defined as
follows:
Gxx (a,σ ) Gxy (a,σ ) 

H ( a, σ ) =   (3)
Gxy (a,σ ) G yy (a,σ )
where Gxx(a, σ), Gxy(a, σ) and Gyy(a, σ) are the convolutions of the Gaussian second order
partial derivatives with the image I in point a respectively.
To minimize computation time, Gaussian is approximated as a set of box filters which
denote the lower scale to compute the blob response maps which are represented by Dxx(a, σ),
Dxy(a, σ) and Dyy(a, σ). The Hessian Matrix is estimated as:
det( H approx ) = Dxx Dyy − (ω Dxy )2 (4)
where ω represents the weight for the energy conservation between the actual and the approx-
imated Gaussian kernels.
Interest points in an image are found at varied scales, where implementation of scale space
is done through an image pyramid. Gaussian smoothing and sub-sampling are used to gener-
ate the pyramid labels.
The SURF descriptor uses the following methodology to find features in an image:
Step I: Setting a reproducible orientation through the use of information from the circular
area around the derived keypoint.
Step II: A square region is constructed, according to the chosen orientation.
867

Step III: Extraction of the SURF descriptor.
For instance, the rotated neighborhood is divided into 16 sub-squares which is further
divided into four squares. The descriptor for sub-square Ai is:
Ai = ( ∑ d x ,∑ d _ x, ∑ d y , ∑ d _ y ) (5)
where dx and dy are HaarWavelet responses in horizontal and vertical directions respectively.
The descriptor emphasizes on the spatial distribution of gradient information inside the
interest point neighborhood.
3.3 Reducing features using PCA

Principal Component Analysis, used for dimensionality reduction, helps to linearly project
high dimensional samples into low dimensional feature space, which help in creating a low
dimensional structure of facial pattern. This methodology drops useless information and
decomposes the facial key points into uncorrelated components.
The major objective of this paper is to reduce the cost, especially for SURF descriptor. A
coarse matching between the key points is obtained by applying PCA on two sets of SURF
features of images. Next, we calculate Kullback-Leibler (KL) divergence similarity score to
improve matching accuracy. Experimental results of our proposed technique concludes that
Figure 2. (a) applying bilateral filtering to the input image-, (b) finding key-points using SURF-PCA,
(c) applying Delaunay triangulation in (b), (d) matching faces with the dataset.
868

it can reduce the dimension of SURF features and the related matching cost in contrary to
the conventional approach by approximately the same precision.
3.4 FLANN
Once facial key points are found, a matching algorithm is used to search in the dataset.
There exist various approximating methods for computing nearest neighbors like brute force
search, Support Vector Machines, FLANN (Muja and Lowe (2009)). They are typically used
for computing the top k-nearest neighbors efficiently ignoring the rank of dataset. To com-
pute short list of nearest neighbors we use FLANN implementation of k-d tree algorithm.
Rather than considering all neighbors in a Rank-Order clustering as given below:
Oa ( b )
d ( a, b ) = ∑ O ( f (i ))
i =1
b a (6)
where fa(i) is the ith face in the neighbor list of a, and Ob(fa(i)) gives the rank of face fa(i) in
face bs neighbor list.
We only use sum of the top k-neighbors where their presence or absence on the short list
is considered more significant than the numerical rank. A distance measure, by summing the
presence or absence of shared nearest neighbors rather than ranks is given by the following
distance function:
min(Oa ( b ),k )
d m ( a, b ) = ∑i =1
I b (Ob ( fa (i )), k ) (7)
where indicator function,
 0, if x in b′ s top nearest neighbor.

I b ( x, k ) =  (8)
 1, otherwise.
3.5 Delaunay triangulation

Let P be the set of keypoints of an image which is sufficiently dense, then a good approxima-
tion is contained in Del(P). This means that an ample subset of the triangular surface of the
Delaunay Triangulation are selected to get an accurate representation of the image. Since a
triangle has only three vertices only 6 key points are matched for a pair of images in a view-
point. Through this, we can achieve robust matching.
We assume projective transformation between the two input images, as given below:
 xh  x 
 yh  = H  y  (9)
   
 k   1 
 h11 h12 h13 

H =  h21 h22 h23  (10)
 h31 h32 h33 
Computing x’, y’ that is non-homogeneous coordinates as:
xh h11x + h12 y + h13

x′ = =
k h31x + h32 y + h33
(11)
yh h21x + h22 y + h23
y′ = =
k h31x + h32 y + h33
869

Table 1. Accuracies of proposed model on different datasets.
Dataset PCA min-Hessian Accuracy
FACE95 0.8 565 75.2

FACE96 0.8 565 76.8
GRIMACE 0.8 565 79.1
ORL 0.8 565 84.4
where (x′, y′) ⇔ (x, y) are pixel-point correspondences and H is Homography transformation
matrix.
The symmetric transfer error is calculated using Euclidean distance and transformation
matrix, d(x, H−1x′)2 + d(x′, Hx)2, for matching keypoints in a pair of image. The inlier’s having
values lesser than threshold are counted.
4 DATASET
To test the robustness and effectiveness of the proposed model we use the FACES95,
FACES96, the ORL Database of Faces and GRIMACE which show variance in pose, expres-
sion, rotation and illumination altogether.
• FACES95 contain 1440 images of 72 individuals; 20 images each. The data set was con-
structed by taking snapshots of 72 individual with a delay of 0.5 seconds between suc-
cessive frames in the sequence. Significant head movement variations was introduced
between images of the same individual. Similar to the above methodology FACES96 was
constructed for 3040 images.
• GRIMACE consist 360 images of 18 individuals. The images taken are variant to scale,
lighting and position of face in the image. In addition the subject made grimaces after
moving his/her head which gets extreme towards the end of the sequence.
• The ORL Database contain images of 40 individuals with total number of images equal to
400. The pictures were taken at different time, light conditions, facial expression and facial
details. The database was utilized in a face recognition project carried out in an association
with the Speech, Vision and Robotics Group of the Cambridge University Engineering
Department.
5 RESULTS AND CONCLUSION
We executed the proposed model over every group of images present in the four benchmark
databases, namely, FACES95, FACES96, GRIMACE and ORL. The results obtained using
our technique on the four datasets are tabulated in Table 1. The table compares accuracy of
our model over different datasets. The retained variance for PCA is set to 0.8 which implies
20% of the total variance is retained. Threshold for min-Hessian value is set to 565 which
accepts the most salient keypoints. The ORL database which is variant to pose, expression
and scale performed well using our method. The other three data sets which in addition to
pose, expression and scale are also variant to illumination which perform moderately low
on our proposed approach as compared to ORL. So the proposed model fails for illumina-
tion and hence should not be used if they are in varied lightning conditions. The usage of
FLANN in our model also increases the speed of matching the images from the database.
REFERENCES
Al-Shabi, M., W.P. Cheah, & T. Connie (2016). Facial expression recognition using a hybrid cnn-sift
aggregator. arXiv preprint arXiv:1608.02833.
870

Bay, H., A. Ess, T. Tuytelaars, & L. Van Gool (2008). Speeded-up robust features (surf). Computer vision
and image understanding 110(3), 346–359.
Bebis, G., T. Deaconu, & M. Georgiopoulos (1999). Fingerprint identification using delaunay triangula-
tion. In Information Intelligence and Systems, 1999. Proceedings. 1999 International Conference on,
pp. 452–459. IEEE.
Dou, J. & J. Li (2012). Robust image matching based on sift and delaunay triangulation. Chinese Optics
Letters 10(2102), S11001.
Dreuw, P., P. Steingrube, H. Hanselmann, H. Ney, & G. Aachen (2009). Surf-face: Face recognition
under viewpoint consistency constraints. In BMVC, pp. 1–11.
Geng, C. & X. Jiang (2009a). Face recognition using sift features. In Image Processing (ICIP), 2009
16th IEEE International Conference on, pp. 3313–3316. IEEE.
Geng, C. & X. Jiang (2009b). Sift features for face recognition. In Computer Science and Information
Technology, 2009. ICCSIT 2009. 2nd IEEE International Conference on, pp. 598–602. IEEE.
Kalantidis, Y., L.G. Pueyo, M. Trevisiol, R. van Zwol, & Y. Avrithis (2011). Scalable triangulationbased
logo recognition. In Proceedings of the 1st ACM International Conference on Multimedia Retrieval,
pp. 20. ACM.
Lindeberg, T. (1998). Feature detection with automatic scale selection. International journal of computer
vision 30(2), 79–116.
Liu, Y., A. Jourabloo, W. Ren, & X. Liu (2017). Dense face alignment. arXiv preprint arXiv:1709.01442.
Lowe, D.G. (2004). Distinctive image features from scale-invariant keypoints. International journal of
computer vision 60(2), 91–110.
Miri, S.S. & M.E. Shiri (2012). Star identification using delaunay triangulation and distributed neural
networks. International Journal of Modeling and Optimization 2(3), 234.
Mostaghim, M., E. Ghodousi, & F. Tajeripoor (2014). Image smoothing using non-linear filters a com-
parative study. In Intelligent Systems (ICIS), 2014 Iranian Conference on, pp. 1–6. IEEE.
Muja, M. & D.G. Lowe (2009). Fast approximate nearest neighbors with automatic algorithm configu-
ration. VISAPP (1) 2(331–340), 2.
Sampath, A., A. Sivaramakrishnan, K. Narayan, & R. Aarthi (2016). A study of household object
recognition using sift-based bag-of-words dictionary and svms. In Proceedings of the International
Conference on Soft Computing Systems, pp. 573–580. Springer.
Shuo, H.,W. Na, & S. Huajun (2012). Object tracking method based on surf. AASRI Procedia 3,
351–356.
Tomasi, C. & R. Manduchi (1998). Bilateral filtering for gray and color images. In Computer Vision,
1998. Sixth International Conference on, pp. 839–846. IEEE.
871

Opportunities and challenges in software defined networking

and network function virtualization
Aiswarya Roy, V.K. Asna, N. Nimisha & C.N. Sminesh

Kerala, India
ABSTRACT: Software Defined Networks (SDN) and Network Function Virtualization

(NFV) are the two emerging paradigms in networking. The control plane is physically sepa-
rated from the forwarding plane and logically centralized in the SDN architecture. SDN
overcomes many limitations in traditional network infrastructures by separating the net-
work’s control plane from the routers and switches. With the decoupling of control plane and
data plane, the entire network is controlled by a centralized controller and network switches
become simple forwarding devices. NFV is the initiation to give a virtualized platform to the
network which is presently carried out by proprietary hardware. The NFV concept was intro-
duced to increase the feasibility and scalability of networks. This paper mainly focuses on the
research opportunities and challenges in the control plane and data plane in SDN and NFV.
1 INTRODUCTION
Software Defined Networking (SDN) is an emerging network architecture. One distinct

change in SDN compared with traditional networking is that the control plane is removed to
a set of controllers (Akyildiz et al., 2014). For small networks, a single centralized controller is
used. But when the size of the network increases a single controller is not enough (Vissicchio
et al., 2015). Multiple controllers are used to solve scalability and reliability issues. One of the
key problems in SDN is the placement of controllers as this could have an impact on network
performance and cost (Braun & Menth, 2014). Another problem in the control plane is the
load balancing among controllers. The data plane uses forwarding devices for the processing
and delivery of packets. The data plane enables data transfer to and from clients.
The NFV concept was introduced to increase the feasibility and scalability of network. In the
traditional networking model each network function is implemented by different physical nodes.
By virtualizing a single physical node can be used for different functions. These virtual function
nodes are known as Virtualized Network Function (VNF), which may run on a single or a set
of Virtual Machines (VMs). Increased elasticity, increased service agility, improved operational
simplicity and faster innovation are the major advantages of NFV. SDN and NFV technologies
are synergistic and they could offer improved programmability and faster service enablement.
This paper is structured as follows: Section 2 discusses the research and challenges in the
control plane; Section 3 discusses the data plane in SDN; Section 4 discusses NFV and the
promising combination of SDN and NFV; and Section 5 concludes the survey.
2 CONTROL PLANE IN SDN
In SDN, the control plane is decoupled from the data plane. It removes the control plane
from the hardware and moved to a set of controllers. The control plane takes the decision
about where and how to forward packets. The small networks need only a centralized con-
troller to work properly. But when the size of the network increases, a single controller is not
873

enough. So multi-controller deployment is used to solve the scalability and reliability issues
raised by the centralized architecture of SDN.
When using multiple controllers, challenges like the controller placement need to be
addressed. Ksentini et al. (2016) proposed a methodology to find the optimal placement of
SDN controllers. In this method the main focus is on the following performance metrics:
The communication overhead and latency between controllers and switches
The communication overhead and latency between controllers
The load balancing between controllers
Objectives of this approach are optimizing the performance of the control plane and
to reduce the cost. This method helps to minimize latency and communication overheads
among controllers and ensure load balancing.
Galich et al. (2017) proposed ways to reduce control traffic latency in an OpenFlow-based
SDN. They derived an algorithm for processing ARP packets using an OpenDaylight con-
troller. The algorithm computes the number of OpenFlow control traffic packets. The total
control traffic latency is dependent on latencies occurring at switches and the controller and
the number of service traffic packets in the network. The main idea proposed is that if the
total control traffic latency exceeds the standard values per approved regulations, it can be
reduced by reducing either switch or controller latency. Finding the location of controllers
and switches is the main challenge in this methodology.
Han et al. ( 2016) proposed an algorithm which optimizes minimum-control-latency based
on greedy controlling pattern design. The latency measurement will not be precise enough
when measuring only based on ping results. In this paper, a methodology is proposed based
on active latency measurement to obtain more accurate control latency. The main objective
of this methodology is to obtain a partition with the optimal number of controllers while
minimizing the control latency. A minimum-control-latency optimized algorithm and a con-
trol latency bounded algorithm are used in this methodology. Minimum-control-latency opti-
mizing can improve the imbalance when partitioning SDN domains. However, to consider
hundreds of topologies in the internet will be a challenge to this methodology.
Load balancing among controllers is another challenge that arises due to multi-controller
deployment. ZHou et al. (2017) propose an effective load balancing mechanism based on a
switches group. The mechanism not only balances the load among controllers, but also solves
the load oscillation and improves time efficiency. A switches selection algorithm and a target
controllers selection algorithm are used for this purpose. The load measurement component
runs on every local controller to measure the load of the controller and periodically sends load
information to the super controller. The current load press of a controller can be calculated as:
ci
pi = (1)
cimax
Table 1. Opportunities and challenges in SDN control planes.
Literature on control
plane Opportunities Challenges
(Ksentini et al. 2016) Optimizing the performance of control Controller placement problem
plane
(Galich et al., 2017) The total control traffic latency can be Finding the location of
reduced by reducing either switch controllers and switches
or controller latency
(Han et al., 2016) Minimize the control latency with Considering hundreds of
optimal number of controllers topologies
(Zhou et al., 2017) Balancing among controllers Proper arrangement of
controllers and switches
874

where ci is the current load of the controller i and cimax is the maximum load capability of the
controller i.
From the above literature it is observed as given in Table 1.
One of the key problems in SDN is the placement of controllers as this could have an
impact on network performance and cost. Therefore, they should be placed in a way to mini-
mize control traffic latency among the switches and the controllers and load balance among
controllers. These observations are shown in Table 1.
3 DATA PLANE IN SDN
The data plane is the part of the network that carries user traffic. The data plane enables data
transfer to and from clients, handling multiple conversations through multiple protocols, and
manages conversations with remote peers. Data plane traffic travels through routers. It uses
forwarding devices for the processing and delivery of packets.
Bozakov and Rizk (2013) propose that in SDN applications switches with various capaci-
ties for control message processing cause unpredictable delays due to their concurrent opera-
tion. This methodology uses a queuing model to characterize the service of a switch’s control
interface to concentrate on this issue. To improve predictability in terms of expected delays,
it implements the control connection to the switch and also enables applications to easily
adapt to the control message processing rates. To transmit controller messages to a specific
switch over an established control connection controller interface typically implemented as
function, send to switch. Therefore, the developer can make sure that a control message has
been accepted by the interface or not. And if a switch is operated over its processing limits,
instead of allowing the messages to queue up at the switch, the socket will get blocked at the
application level. As a result, the application may adjust its sending rate much quicker. The
current SDN abstraction model is maintained which is the main benefit of this methodology.
Pontarelli et al. (2017) present a method to implement complex tasks into stateful SDN pro-
grammable data planes. The presented method proposes to use the internal microcontroller,
typically used to configure the programmable data plane, to also perform some complex oper-
ations that do not require to be executed on each packet. These operations can be executed
on a set of data gathered by the data plane and processed in a time scale that is much higher
than the time window of a packet, but is much less than time scale needed for an external SDN
controller. Moreover, the use of the configuration microcontroller instead of an external SDN
controller to avoid the exchange of data on the control links permits a fine grain tuning of
the operations to perform from the timing point of view. Extending the original stateless SDN
data plane paradigm enables the execution of simple control functions directly into the fast
path. In this paper, a different approach is to extend stateful SDN data plane with the support
of a set of lazy operations. The term lazy refers to a set of operations that are triggered by the
reception of a packet inside the pipeline, but return its result after a certain amount of time
without blocking the forwarding of the packet itself inside the switch.
Zhang et al. (2014) proposed Big Switch abstraction as a specification mechanism for
high-level network behavior. This specification allows the operator to define end-to-end flow
policies. This can be used for placing rules on individual switches by the network operating
system. This is forced to do so by the limited capacity of the Ternary Content Addressable
Memories (TCAMs) used for rules in each switch. Using a centralized rule management sys-
tem, it is compiled down to individual rules on different switches. Permitted (PERMIT rule)
or dropped (DROP rule) are the rules related to each packet. For placing rules on switches
the authors proposed a solution based on Integer Linear Programming (ILP). While optimiz-
ing the number of rules and maintaining the switch capacity constraints, this can be applied
on a given firewall policy. Switch priority, capacity, and policy constraints are satisfied by
ILP. And this also optimizes certain objective functions such as minimizing the total number
of rules. But complex rule placement constraints such as monitor certain packets, that do
not want to let firewall rules to block packets before they reach the monitoring rules is not
supported by this concept.
875

Table 2. Opportunities and challenges in SDN data plane.
Literature on
data plane Opportunities Challenges
(Pontarelli Because almost all pro grammable Data To implement complex tasks into
et al., 2017) planes already haves an internal a stateful SDN programmable
micro-controller used to configure the data plane.
memory tables, method proposed in
the paper could be widely applied.
(Zhang et al., Applicable to real-sized net works. The meaning of the original
2014) policies have to maintained
by the results obtained from
this method.
(Bozakov & Extension of this method can manage Device hetero—geneity as an
Rizk, 2013) more complex control net work inherent property of SDN
topologies and distributed controllers. which must be considered.
From the above literature it is observed that the key challenges in SDN data plane are
implementation of complex tasks into a SDN programmable data plan and sharing of rules
across different paths. Proper handling of these problems leads to better opportunities. These
observations are shown in Table 2.
4 NETWORK FUNCTION VIRTUALIZATION
The NFV concept was introduced to make the network more flexible and simpler than the
traditional network concept which is dependent on hardware constraints (Lopez, 2014). NFV
is about providing network functions by software rather than using a number of hardware.
Most importantly, the benefit of NFV is the flexibility to easily, rapidly, and dynamically
provision and instantiate new services in various locations.
SDN is another network paradigm. The combination of both SDN and NFV started a new
era of networks. The combination of SDN principles and NFV infrastructure can change
the way of building, deploying and controlling network applications, which is described by
King et al. (2015). To provide flexibility, scale, distribution and bandwidth to match the user
demands, flexible and dynamic optical resources are used in the current transport networks.
Due to the requirement of significant engineering resources, these are non-real-time capabili-
ties and often lack the flexibility for dynamic scenarios. To overcome these challenges, Daniel
et al. propose a new architecture with the combination of SDN, NFV, flexi-grid, and ABNO
(Application-Based Network Operations). This new architecture has the capability to deploy
a vCDN (virtualized Content Distribution Network) which is capable of scaling the user
bandwidth demand and control resources programmatically.
Another combination of SDN and NFV with IoT is proposed by Ojo et al. (2016). A typi-
cal Io-T architecture has a 3-layered structure, which are the perception layer, the network
layer and the application layer. By adopting the SDN concept in this Io-T architecture the
network layer is divided into two planes; control plane and data plane. The data layer com-
prises SDN routers and switches to forward packets. These switches and routers in the data
layer are programmatically controlled in the control layer. By enabling NFV on this new
SDN adopted IoT architecture, the network agility and network efficiency of IoT applica-
tions increases. The IoT—gateway becomes dynamic, scalable and elastic by virtualization.
Virtualization makes the infrastructure more flexible and sustainable by decoupling the net-
work control and management function from the hardware.
The future internet scenario named virtual presence, by leveraging on the joint SDN/NFV
paradigm, together with a fog computing approach is proposed by Faraci and Lombardo
(2017). Virtual presence is achieved by exporting a real hardware or software resource, that
876

Table 3. Opportunities and challenges of NFV.
Literature on NFV Opportunities Challenges
(King et al., 2015) Capable to respond to high bandwidth real NFV environment is
time and predicted video stream demands more dynamic than
the traditional one
(Ojo et al., 2016) SDN and NFV concepts help to solve new Security issues of NFV
challenges of IoT
(Faraci & Smart device sharing service could increase Having toco exist in
Lombardo, 2017) the resource utilization a cloud in tegrated
environment
is physically or virtually located in a personal network of a given user such as a smart TV

or a virtual set-top box, to another personal network. The proposed architecture allows a
virtual presence to be instantiated in a personal network, in such a way that it allows using
it as present locally in the visited personal network. In this proposed architecture, SDN and
NFV concepts help to run both network and application functions in a virtualized form on
the nodes at the edge of the network.
A comparison of SDN and NFV for redesigning the LTE (Long Term Evolution) packet
core is done by Jain et al. (2016). The experiment result shows that an NFV based implemen-
tation is better suited for networks with high signaling traffic, because handling the com-
munication with the SDN controller quickly becomes the bottleneck at the switches in the
SDN-based EPC (Evolved Packet Core). On the other hand, an SDN-based EPC is a better
choice when handling large amounts of data plane traffic, because SDN switches are often
more optimized for packet forwarding than virtualized software appliances.
Along with the opportunities and benefits, NFV poses significant security problems. The
security issues in Virtualized Network Functions (VNFs), which is an important part of
NFV’s architecture is discussed by Aljuhani and Alharbi (2017). These security issues are
classified into three: insider attack, outsider attack, and between VNFs attack.
The combination of SDN and NFV provides vast opportunities in networking. It could
make the transport network more dynamic and feasible as described by King et al. (2015).
Adopting SDN and NFV concepts in IoT could provide solutions to address the challenges
in IoT (Ojo et al., 2016). The virtual presence concept could provide a lot of opportunities
(Faraci & Lombardo, 2017). Along with a lot of opportunities, NFV also faces challenges.
These opportunities and challenges are shown in Table 3. The major problems in dealing with
NFV are the security issues.
5 CONCLUSION
The paper mainly focuses on the research opportunities and challenges in SDN and NFV.
From conducting the literature survey on the control plane and the data plane in SDN, iden-
tified research opportunities and challenges are summarized. NFV provides how network
functions can be done using software rather than hardware. It is observed that a combination
of NFV and SDN is really a promising technology to solve many of the complex networking
issues in the existing network architectures.
REFERENCES
Akyildiz, I. F., Lee, A., Wang, P., Luo, M. & W. Chou, W. (2014). A roadmap for traffic engineering in
SDN-Openflow networks. Computer Networks, 71, 1−30.
Aljuhani, A. & Alharbi, T (2017). Virtualized network functions security attacks and vulnerabilities.
In IEEE 7th Annual Computing and Communication Workshop and Conference (CCWC) (pp. 1–4).
New York, NY: IEEE.
877

Bozakov, Z. & Rizk, A. (2013). Taming SDN controllers in heterogeneous hardware environments.
In Second European Workshop on Software Defined Networks (EWSDN) (pp. 50–55). New York, NY:
IEEE.
Braun, W. & Menth, M. (2014). Software-defined networking using Openflow: Protocols, applications
and architectural design choices. Future Internet, 6(2), 302–336.
Faraci, G. & Lombardo, A. (2017). An NFV approach to share home multimedia devices. In IEEE
Conference on Network Softwarization (Net-Soft) (pp. 1–6). New York, NY: IEEE.
Galich, S., Deogenov, M. & Semenov, E. (2017). Control traffic parameters analysis in various software-
defined networking topologies. In International Conference on Industrial Engineering, Applications
and Manufacturing (ICIEAM) (pp. 1–6). New York, NY: IEEE.
Han, L., Li, Z., Liu, W., Dai, K. & Qu, W. (2016). Minimum control latency of SDN controller place-
ment. In IEEE Trustcom/BigDataSE/ISPA (pp. 2175–2180). New York, NY: IEEE.
Jain, A., Sadagopan, N., Lohani, S.K. & Vutukuru, M. (2016). A comparison of SDN and NFV for
redesigning the LTE packet core. In IEEE Conference on Network Function Virtualization and Soft-
ware Defined Networks (NFV-SDN) (pp. 74–80). New York, NY: IEEE.
King, D., Farrel, A. & Georgalas, N. (2015). The role of SDN and NFV for flexible optical networks:
Current status, challenges and opportunities. In 17th International Conference on Transparent Optical
Networks (ICTON) (pp. 1–6). New York, NY: IEEE.
Ksentini, A., Bagaa, M., Taleb, T. & Balasingham, I. (2016). On using bargaining game for opti-
mal placement of SDN controllers. In IEEE International Conference on Communications (ICC)
Lopez, D.R. (2014). Network functions virtualization: Beyond carrier-grade clouds. In Optical Fiber
Communications Conference and Exhibition (OFC) (pp. 1–18). New York, NY: IEEE.
Ojo, M., Adami, D. & Giordano, S. (2016). A SDN-IoT architecture with NFV implementation.
In IEEE Globecom Workshops (GC Wkshps) (pp. 1–6). New York, NY: IEEE.
Pontarelli, S., Bruschi, V., Bonola, M. & G. Bianchi, G. (2017). On offloading programmable SDN
controller tasks to the embedded microcontroller of stateful SDN data planes. In IEEE Conference
on Network Softwarization (NetSoft) (pp. 1–4). New York, NY: IEEE.
Vissicchio, S., Tilmans, O., Vanbever, L. & Rexford, J. (2015). Central control over distributed routing.
ACM SIGCOMM Computer Communication Review 45(4), 43–56.
Zhang, S., Ivancic, F., Lumezanu, C., Yuan, Y., Gupta, A. & Malik, S. (2014). An adaptable rule place-
ment for software-defined networks. In 44th Annual IEEE/IFIP International Conference on Depend-
able Systems and Networks (DSN) (pp. 88–99). New York, NY: IEEE.
Zhou, Y., Wang, Y., Yu, J., Ba, J. & Zhang, S. (2017). Load balancing for multiple controllers in SDN
based on switches group. In 19th Asia-Pacific Network Operations and Management Symposium
(AP-NOMS) (pp. 227–230). New York, NY: IEEE.
878

A parallel framework for maximal clique enumeration
Thilak Anusree & K. Rahamathulla

Kerala, India
ABSTRACT: Graphs play a major role in modeling many real-world problems. Due to the
availability of huge data, the graph processing in serial environment become more complex.
Thus, fast and efficient algorithms which work effectively utilizing the modern technologies
are required. A maximal clique problem is one of the graph processing methods which is
used in many applications. The Bron-Kerbosch (BK) algorithm is the most widely used and
accepted algorithm for listing each and every maximal clique in a graph. Here an idea of a
parallel version of BK algorithm is proposed which will reduce the computation time to a
large extent than its serial implementation. It utilizes the cluster computing strategy.
1 INTRODUCTION
1.1 Maximal clique enumeration problem

The listing of all maximal cliques which are present in a graph is called the maximal clique
enumeration problem. Let G be a graph, where V and E are the set of vertices and edges of
G respectively. A clique in graph G is a subset of vertices of G where each and every vertex
belonging to the subset is connected to all other vertices within the subset.
1.2 BK algorithm
Of all the algorithms for maximal clique enumeration, the Bron-Kerbosch (BK) algorithm is
the most widely used method. Designed by two Dutch scientists, Joep Kerbosch and Coen-
raad Bron, this algorithm was published in 1973. However, although many different algo-
rithms, which are theoretically better than BK on the inputs, have been proposed, BK still
remains better in practical applications for listing all the cliques.
1.3 Parallel approach

In real-world scenarios, the sizes of graphs are so large that the number of nodes range from
hundreds to thousands. The processing of such large graphs using a serial algorithm may
consume a lot of time.
Because of this higher time complexity of serial algorithms, a parallel approach can be
used to overcome this difficulty and to give the result in much less time. Here we try to imple-
ment a parallel version of a BK algorithm which will efficiently enumerate all cliques in any
given graph that are maximal.
1.4 Graph based applications

Structure alignment of 3-D protein, gene expression analysis, social hierarchy detection, and
genome mapping are some of the applications which use graph processing. Maximal clique
enumeration is the main method which has been used in these applications.
879

Figure 1. Maximal cliques in a graph.
1.5 Organization of paper

This paper is being organized as follows: the background works are discussed in Section 2;
Section 3 gives the literature survey where some approaches are discussed; Section 4 outlines
the proposed method; And Section 5 details the conclusion.
2 BACKGROUND
A maximal clique is a completely connected subgraph which is not the subset of any other
bigger clique in that graph, i.e. we can never expand an already existing clique (which is maxi-
mal in nature) by adding one more neighboring vertex to it. Every graph having n vertices can
have a maximum of 33 maximal cliques (Tomita et al., 2006).
The biggest clique in a graph is the maximum clique. The maximum common subgraph
problem can be reduced to maximal clique enumeration problem (Jayaraj et al., 2016) where
the former is an NP-complete problem. Figure 1 shows maximal cliques in an example graph G.
The maximal clique enumeration problem originally arose in different areas of research
as a set of related problems. The algorithms for solving those issues can be regarded as the
initial algorithms for clique detection. The algorithm of Harary and Ross, (1957) was the
first broadly acknowledged endeavor for listing all of the cliques which are maximal in a
graph. which displays a technique for discovering connections between individuals utilizing
the sociometric information between them which will form a clique.
3 LITERATURE SURVEY
With the introduction of the backtracking search strategy, the maximal clique enumeration
problem gained momentum to a great extent. The efficiency of combinatorial optimization algo-
rithms is improved by using backtracking which limits the search space size of the algorithm.
A set of viability criteria is established that is used to help the backtracking algorithm from
explorating the non-promising paths (Schmidt et al., 2009).
3.1 Maximal clique enumeration methods

The BK algorithm lists all the cliques present in a graph exactly once (Jayaraj et al., 2016).
It uses three sets for finding maximal cliques using recursion. It uses an approach in which
only the nearest neighbors of the node are considered. A node is taken and its neighbors are
explored recursively to find whether they form a clique or not. The BK algorithm is given in
Algorithm 1 (Bron & Kerbosch, 1973).
880

Wen et al. (2017) propose a Multi Objective Evolutionary Algorithm (MOEA) based on
maximal cliques to detect the overlapping community. In comparison with existing systems,
the MOEA has low computational cost and high partition accuracy for detecting the struc-
ture of overlapping communities. Both the efficiency and effectiveness of MOEA was vali-
dated using real-world and synthetic networks.
3.2 Parallel methods for solving clique problems

In a large graph to enumerate all the maximal bi-cliques, Mukherjee and Tirthapura (2017)
proposed a technique using MapReduce. The enumeration is parallelized by using the basic
clustering framework. This is followed by optimizations one each for load balancing improve-
ment and redundant work reduction. The algorithm was found be effective in handling large
graphs in the experimental results. This is the first successful work to list all the bi-cliques
using a parallel implementation in large graphs. The previous works were on smaller graphs
which used mainly sequential methods for the enumeration of bi-cliques.
Xu et al. (2016) first proposed a distributed algorithm for the computation of cliques which
are maximal. As the vertices in a graph can have high degrees the problem of skewed work-
loads can arise. This problem has been efficiently addressed here. Thus, in the case of real-world
graphs, the time complexity in the worst case for computing the cliques which are maximal has
been reduced effectively. Next, to analyze efficiently the maximal cliques set, they proposed
some fundamental algorithms to process the fundamental query operations. In a case were the
graph is being updated, algorithms were devised to efficiently update the maximal cliques set.
The efficiency of the algorithms was verified using real-world graphs under different
domains for finding, querying, and updating the maximal cliques set.
Figure 2. System design.
881

4 PROPOSED METHOD
Here we would like to implement a new parallel version of the already existing BK algo-
rithm, which will be efficient in maximal clique enumeration. Here we take each node in the
graph and explore its neighbors in a parallel manner, i.e. each node will be explored by an
independent worker node thus reducing the overall computation time. Here we have a cluster
manager who divides the work between the worker nodes. The worker nodes, which are many
in number, work simultaneously to produce a faster result.
In the proposed system, the graph is stored in a file as an edge list, i.e., each edge will be
represented individually by a set of the two vertices it connects. The graph will be processed
using branch and bound criteria and a tree will be created. Each worker node will traverse the
tree in a bfs manner to find the cliques in it. Later, the result from all the worker nodes will
be combined to form the list of all maximal cliques.
Here we can implement this by using any parallel framework such as graph X in Apache
Spark, or by using CUDA. Graph X is built on top of Apache Spark as an embedded frame-
work and is used for processing graphs. It’s a distributed dataflow system which is widely
used. The sufficient graph abstraction that is required to express existing graph APIs is being
provided by graph X (Gonzalez et al., 2014).
CUDA is the parallel computing architecture of NVIDIA that by harnessing the power of
the graphics processing unit enables the users to have an increased computing performance.
This can also be used to implement an efficient parallel version of BK algorithm.
5 CONCLUSIONS
Many real-world problems can be modeled using graphs. And, maximal clique enumeration
problem has a wide range of applications in the fields of drug discovery and its analysis,
social hierarchy detection and many more. The serial computation of maximal cliques is a
time-consuming task when the sizes of graphs are large and we know that the real-world
graphs are mostly large in size. Thus, here we propose a parallel algorithm for the already
existing BK algorithm for the processing of large graphs. The BK algorithm is one of the
most widely accepted graph algorithms for listing all the maximal cliques present in a graph.
REFERENCES
Bron, C. & Kerbosch, J. (1973). Algorithm 457: finding all cliques of an undirected graph. Communica-
tions of the ACM, 16(9), 575–577.
Gonzalez, J.E., Xin, R.S., Dave, A., Crankshaw, D., Franklin, M.J. & Stoica, I. (2014). Graphx: Graph
processing in a distributed dataflow framework. In OSDI, Volume 14, (pp. 599–613).
Harary, F. & Ross, I.C. (1957). A procedure for clique detection using the group matrix. Sociometry
20(3), 205–215.
Jayaraj, P., Rahamathulla, K. & Gopakumar, G. (2016). A GPU based maximum common subgraph
algorithm for drug discovery applications. In IEEE International Parallel and Distributed Processing
Symposium Workshops (pp. 580–588). New York, NY: IEEE.
Mukherjee, A. & Tirthapura, S. (2017). Enumerating maximal bicliques from a large graph using
MapReduce. IEEE Transactions on Services Computing.
Schmidt, M.C., Samatova, N.F. Thomas, K. & Park, B.-H. (2009). A scalable, parallel algorithm for
maximal clique enumeration. Journal of Parallel and Distributed Computing, 69(4), 417–428.
Tomita, E., Tanaka, A. & Takahashi, H. (2006). The worst-case time complexity for generating all maxi-
mal cliques and computational experiments. Theoretical Computer Science, 363(1), 28–42.
Wen, X., Chen, W.-N., Lin, Y., Gu, T., Zhang, H. Li, Y., Yin, Y. & Zhang, J. (2017). A maximal clique
based multi objective evolutionary algorithm for overlapping community detection. IEEE Transac-
tions on Evolutionary Computation, 21(3), 363–377.
Xu, Y., Cheng, J. & Fu, A.W.-C. (2016). Distributed maximal clique computation and management.
IEEE Transactions on Services Computing, 9(1), 110–122.
882

Retrieving and ranking similar questions and data-driven answer

selection in community question answering systems
V. Dhrisya & K.S. Vipin Kumar

Kerala, India
ABSTRACT: Different techniques have been utilized to study how to find similar questions
from recorded archives. These similar questions will have multiple answers associated with them
so that end-users have to carefully browse to find a relevant one. To tackle both problems, a novel
method of retrieving and ranking similar questions, combined with a data-driven approach of
selecting answers in Community Question Answering (CQA) systems, is proposed. The presented
approach for similar question retrieval combines a regression procedure that maps topics deter-
mined from questions to those found from question-answer pairs. Applying this can avoid issues
due to distinctions in vocabulary used within question-answer sets and the inclination of queries
to be shorter than their answers. To alleviate answer-ranking problems, a scheme via pairwise
comparisons is presented. In the offline learning component, the scheme sets up positive, negative,
and neutral training samples represented as preference pairs by means of data-driven perceptions.
The model incorporates these three sorts of training samples together. At that point, utilizing the
offline prepared model, the answer candidates are sorted to judge their order of preference.
1 INTRODUCTION
One of the fastest developing customer-generated content portals, the Community Question
Answering (CQA) system has emerged as a huge market that satisfies complex information
needs. CQA provides a platform for customers to ask questions on any topic and also answer
others as they wish. It also enables a search through the recorded past Question-Answer (QA)
set. Conventional factual QA can be answered by simply retrieving named entities or content
from available documents, while CQA extends its significance to answer complicated ques-
tions such as reasoning, open-ended, and advice-seeking questions. CQA places few restric-
tions, if any, on who can post and who can answer a query and is thus quite open. Both the
general CQA sites such as Yahoo! Answers and Quora, and the specialized ones like Stack
Overflow and HealthTap, have had a significant influence on society in the last decade.
Even though there is active user participation, certain phenomena result in question dep-
rivation in CQA portals. For example, users have to wait a long time before getting responses
to their queries and a considerable number of questions never get any answer, and the askers
are left unsatisfied. The situation is probably caused by the following: (1) the posted queries
may be ambiguous, ineffectively stated or may not invoke curiosity; (2) the CQA systems may
not effectively direct recent questions to the appropriate answerers; (3) the potential answer-
ers, having the required knowledge, may not be available or are overwhelmed by the number
of incoming questions. This third situation often arises in specialized CQA portals, where
answering is restricted to authorized specialists only. With reference to the first case, ques-
tion quality modeling can check the question quality and can assist in requesting that askers
restructure their queries. For the other two cases, the situation can be addressed by means of
question routing. Question routing is performed by expertise matching and consideration of
the likelihood of potential answerers. It works by exploring the human resources currently
associated with the system. Besides that, solved past queries can be reused to answer newly
883

presented ones. As time goes on, an enormous set of QA pairs are recorded in databases.
New questions have an increased chance of directly obtaining a relevant answer by brows-
ing from the archives, in lieu of waiting for a while to get a fresh response. This proposal has
attracted the attention of many researchers who work to make the automatic extraction of
recorded, relevant information from the CQA database possible.
Taking into account the QA structure and its meaning, questions or answers with similar
meaning may not be lexically similar, and this is defined as a ‘lexical chasm’. Consider an
example: the questions such as “Where can I learn python on the internet for free?” and
“Are there any sites for learning python?” mean the same thing, that is, they are semantically
similar but lexically different. The reverse case can also arise where questions have words in
common but imply varied meanings. Beyond the requirement for precisely determining a
question’s meaning, the solution should also handle spell checking, uncertainty avoidance,
and short questions. Similar questions can be obtained by correlating the newly presented
question to the meaning of recorded questions, because many previous works show that it is
inefficient to find similar questions based only on their answers. Recently, this problem has
been related to the topic of modeling because it scales down the dimensionality of textual
information by comparison with conventional techniques such as ‘bag-of-words’ (BOW), and
it handles uncertainty and similarity comprehensively. The contribution of the present work
as a whole can be said to be twofold: first, topics of the questions and answers in the data-
base are modeled by applying Latent Dirichlet Allocation (LDA), and second, the QA topic
distribution of a new question is estimated using a regression step.
2 LITERATURE REVIEW
State-of-the-art techniques such as BOW, with its specific properties, term frequency–
inverse document frequency (tf-idf) and BM25 (Robertson & Walker, 1997), can calculate
lexical similarity between two documents, but they do not consider their semantic and con-
textual information. In the past decade, topic modeling has been used as an important tech-
nique in the field of text analysis. The topics that characterize documents can be treated as
its semantic representation. Therefore, to find the semantic similarity between documents,
we can use topic distributions obtained using LDA. Further, various approaches for apply-
ing topic modeling to historical QA have been proposed. For finding similar questions
to the problem, topic modeling along with topic distribution regression (Chahuara et al.,
2016) is used.
Consider a corpora C of size L with C consisting of many question-answer pairs:
C = {(q1, a1), (q2, a2),…,(qL, aL)}
where Q = {q1, q2,…, qL} and A = {a1, a2,…, aL} ∀ (qi, ai) ∈ C: qi ∈ Q, ai ∈ A are, respectively,
question and answer sets.
Answers compared with questions in such portals are likely to be longer and the questions
may have only limited relevant words. This can restrict a model’s capacity to detect hidden
trends. A way of overcoming this is by inferring each question qi such that it has its text along
with a title and description. Moreover, every qi may be associated with multiple answers and
these answers are concatenated to obtain each term ai. This is done so as to best determine
the question’s relevance based on the contextual details provided by them. Figure 1 illustrates
the proposed framework. The job of retrieving similar questions can be optimized to the task
of ranking the QA pairs contained in the created set C, assigning a similarity to question
q, and generating a result having its top-ranked element as that with the highest similarity
found.
In the learning phase of the task of extracting similar questions, the set C already created is
used in training two topic models: first, LDA on the question set Q; second, LDA on the ques-
tion-answer pair set QA. The learning phase provides topic distributions associated with the
sets Q and QA, θiQ and θiQA as the result. Using these topic distribution samples, a regression
884

Figure 1. An analysis of the system’s learning and deduction.
model is trained to learn the translation function between the Q and QA distributions. Dur-
ing deduction, using the Q set LDA model, we determine the topic distribution of a question
(θ*Q), which is mapped to its probable QA topic distribution (θ*QA) using the trained regression
model. Last, according to the similarity between each pair’s topic distribution and the new
question’s QA topic distribution, a similarity value is calculated and the value is used to rank
the QA corpora. The questions, respective to the QA distribution, that were found similar to
the presented question can be considered as the output of the primary batch of processing.
The answer-selection problem in CQA is analogous to the conventional ranking task,
where the given question and its set of answers are comparable to a query and a set of
relevant entities. The objective is thus optimized to find an ideal ranking system of the
answer candidates according to their pertinence, exactness and quality with respect to
the given query. A ranking function which uses relevance intuition can be designed in
the following three ways: (1) pointwise − in this type of method (Dalip et al., 2013; Shah
& Pomerantz, 2010) the relevance measure of each individual QA pair is estimated by
a standard classification or regression model; (2) pairwise − in these kinds of methods
(Bian et al., 2008; Hieber & Riezler, 2011; Cao et al., 2006; Li et al., 2015), the preference
of two answer candidates is predicted using a 0–1 classifier; (3) listwise − in this type, the
integrated ranking of all candidate answers to the same question is performed at the same
time (Xu & Li, 2007).
The answer selection in the problem is addressed using the novel PLANE model proposed
by Nie et al. (2017). Given a question, it requires a set of top k relevant questions Q = q1,…,qk
from the QA repositories and, according to Nie et al. (2017), this is done using a question-
matching algorithm k-NN. According to this, question qi is assumed to have a set of mi ≥ 1
answers, represented by Ai = a0, a1,…, ami whereby ai0 is the answer of qi selected as best by
community users. From the identified relevant questions, a learning-to-rank design is devel-
oped to sort all the answers associated with them. Two training sets X and U are built from
the set of QA pairs. x1 and x2 denote the N-dimensional feature vectors of the two QA pairs
that are compared in a single comparison. Also, y is denoted as the preference relationship of
x, whose value is found as below:
y = + 1, if x1 > x2
y = − 1, if x2 > x1
885

With these variables, X = {(xi, yi)}Ni = 1 with preferable labels are built. It can be inferred
that this preference relationship does not work with the unimportant answers of a particular
question. To include those answers, it is essential to have:
(qi, aji) ≅ (qi, aki)
where ≅ represents a neutral preference relationship between the two QA pairs under consid-
eration. u(1) and u(2) represent, respectively, the N-dimensional feature vectors of the two QA
pairs. Taking into account all pairs of comparisons in this pattern, U = {(uj,0)}Mj = 1 is created.
Jointly incorporating X and U, the following pairwise learning-to-rank model is proposed:
N M
min ∑ [1 − yi wT x i ] + λ w 1 + µ ∑ wT u j ,
w
i =1 j =1

where xi = xi1 – xi2 ∈ RN and uj = uj1uj2 ∈ RD denote the two training instances from X and
U, respectively; symbols N and M denote, respectively, the number of preference pairs in
X and U, and the desired coefficient vector is represented by w ∈ RN. The first term in the
equation indicates a hinge loss function, which helps in the binary preference judgment job,
and it gives a relatively rigid and convex upper limit on the binary indicator function. The
conventional formulation for the Support Vector Machine (SVM) can be considered equiva-
lent with empirical risk minimization of this loss. Pertaining to the support vectors, points
lying outside the margin boundaries that are properly classified will not be penalized, while
points on the wrong side of the hyperplane or within the margin boundaries will be penalized
in a linear mode, proportional to their distance from the proper boundary. The second term
represents a l1 norm. It helps in feature selection and in regularizing the coefficient value’s
summation that helps in penalizing the preference distance between unimportant answers of
the same question.
3 PROPOSED FRAMEWORK
The system tries to solve the problem of finding similar questions and selecting an answer
from them using the proven efficient methods of the relevant area. For the purpose of train-
ing, we require a large amount of data. The historical archive data from online general CQA
websites such as Quora and Yahoo! Answers, and specialized ones such as Stack Overflow and
HealthTap, are suitable as sources of data for the system. Retrieving this data and processing
it into the form required comprises the first phase of the framework.
The data is then fed to the topic modeling phase using the LDA method. LDA works
well for topic modeling with data from a variety of topics. As the data is from the CQA,
where questions from different fields appear, the LDA should be efficient. After finding the
translation function between questions and question-answer pairs, as described in Section 2,
using the regression model, the topic model of the expected answer of the new question is
determined. Those question-answer pairs that are similar to this found topic are considered
to be “similar questions”.
Now, using these question-answer pairs, the PLANE model is trained in the form of posi-
tive, negative and neutral preference pairs as described in Section 2. The PLANE model is
trained offline using the constructed pairs. On providing the new input question, it returns
the ranked answers. The best answer will be the one that solves the similar question and that
was chosen by users. Finally, the system is evaluated for its performance by comparison with
systems with other techniques. Accuracy, precision and recall are the metrics that can meas-
ure the performance of such learning systems. Practically, user feedback can be collected
from user satisfaction with the related answer shown. On new questions arriving at the site
under the same topic, a trace-back mechanism can increase the efficiency of prediction.
886

Figure 2. Proposed framework flowchart.
4 EVALUATION OF PERFORMANCE
The data required for the system is from the online community question-answering websites.
As it tries to solve a real-time problem of finding a suitable answer at the time a new ques-
tion arrives, the efficiency of the system can be measured primarily from user feedback. The
objectives for which the model was designed can be evaluated to check whether they were
met:
• to return the best existing relevant information from the historical archive data;
• to reduce the waiting time to get answers to the question.
Because the two phases in the system use novel techniques that are said to outper-
form other methods for the same problem, the combination of both should give better
efficiency.
887

5 CONCLUSION
This paper proposes a combination of two techniques to address answer retrieval from
archive data for a new question that arrives. For finding similar questions from the existing
data, topic modeling using LDA and regression models is preferred. In the successive phase,
for ranking answers, an enhanced SVM model named PLANE is used. As mentioned above,
the system uses novel techniques that are proven to outperform other methods, and the pro-
posed system as a whole should provide better results in terms of accuracy.
REFERENCES
Bian, J., Liu, Y., Agichtein, E. & Zha, H. (2008). Finding the right facts in the crowd: Factoid question
answering over social media. In Proceedings of the 17th International Conference on World Wide Web
(pp. 467–476). New York, NY: ACM.
Cao, Y., Xu, J., Liu, T.Y., Li, H., Huang, Y. & Hon, H.W. (2006). Adapting ranking SVM to document
retrieval. In Proceedings of the 29th Annual International ACM SIGIR Conference on Research and
Development in Information Retrieval (pp. 186–193). New York, NY: ACM.
Chahuara, P., Lampert, T. & Gancarski, P. (2016). Retrieving and ranking similar questions from
question-answer archives using topic modelling and topic distribution regression. In N. Fuhr, L.
Kovács, T. Risse & W. Nejdl (Eds.), Research and advanced technology for digital libraries. TPDL
2016. Lecture Notes in Computer Science (Vol. 9819, pp. 41–53). Cham, Switzerland: Springer.
Dalip, D.H., Goncalves, M.A., Cristo, M. & Calado, P. (2013). Exploiting user feedback to learn to rank
answers in Q&A forums: A case study with stack overflow. In Proceedings of the 36th International
ACM SIGIR Conference on Research and Development in Information Retrieval (pp. 543–552). New
York, NY: ACM.
Hieber, F. & Riezler, S. (2011). Improved answer ranking in social question-answering portals. In Pro-
ceedings of the 3rd International Workshop on Search and Mining User-Generated Contents (pp.
19–26). New York, NY: ACM.
Li, X., Cong, G., Li, X.L., Pham, T.A.N. & Krishnaswamy, S. (2015). Rank-geoFM: A ranking based
geographical factorization method for point of interest recommendation. In Proceedings of the 38th
International ACM SIGIR Conference on Research and Development in Information Retrieval (pp.
433–442). New York, NY: ACM.
Nie, L., Wei, X., Zhang, D., Wang, X., Gao, Z. & Yang, Y. (2017). Data-driven answer selection in com-
munity QA systems. IEEE Transactions on Knowledge and Data Engineering, 29(6), 1186–1198.
Robertson, S.E. & Walker, S. (1997). Some simple effective approximations to the 2-Poisson model for
probabilistic weighted retrieval. Readings in Information Retrieval, 345, 232–241.
Shah, C. & Pomerantz, J. (2010). Evaluating and predicting answer quality in community QA. In Pro-
ceedings of the 33rd International ACM SIGIR Conference on Research and Development in Informa-
tion Retrieval (pp. 411–418). New York, NY: ACM.
Xu, J. & Li, H. (2007). AdaRank: A boosting algorithm for information retrieval. In Proceedings of
the 30th Annual International ACM SIGIR Conference on Research and Development in Information
Retrieval (pp. 391–398). New York, NY: ACM.
888

Convolutional Neural Network (CNN) framework proposed

for Malayalam handwritten character recognition system
using AlexNet
J. Manjusha
Department of Computer Science and Engineering, Government Engineering College, Thrissur, India
A. James
Department of Computer Science and Engineering, Government Engineering College, Thrissur, India
Saravanan Chandran
National Institute of Technology, Durgapur, West Bengal, India
ABSTRACT: Handwritten character recognition has wide application in today’s scenarios

such as bank cheque k processing and tax returns. The phases in character recognition are image
scanning, noise removal, segmentation, extracting features, and classification. Different methods
are used for recognizing characters in different languages. Most of them use handcrafted feature
extraction methods which are time-consuming. Here we use automatic feature extraction using
a Convolutional Neural Network (CNN) which take less time for training and testing of data
and produces more error-free results compared to other methods. Very few works on character
recognition using CNN have been reported to have a high accuracy of prediction. The model
used here is AlexNet architecture which considerably reduces time and errors in predicting the
output. Compound Malayalam character recognition is also focused on in this work using CNN.
1 INTRODUCTION
Optical Character Recognition (OCR) is one of the challenging areas of computer vision
and pattern recognition. It is the process of converting text or handwritten documents into a
scanned and digitized form for easy recognition. It is categorized as offline and online based
on the method by which image acquisition is carried out. Malayalam handwritten character
recognition is very important because of its use by a large population of people but the vari-
ability in writing style and representation make it difficult for recognition. Some of the appli-
cations of OCR is data entry for business applications like cheques processing and passport
verification. Handwritten character recognition includes like image acquisition, preprocessing,
segmentation of characters, feature extraction, and recognition. Compared to all other stages,
feature extraction plays an important role in determining the accuracy of recognition. Some
of the challenges faced by handwritten Malayalam character recognition systems includes the
un-availability of a standard data set and the unlimited variations in human handwriting. The
traditional methods usually require artificial feature design and manually tuning of the classi-
fier. Particularly, the performance of the traditional method can be determined to a large extent
by using empirical features. The traditional method has reached its limit through decades of
research while the emergence of deep learning provides a new way to break this limit. In this
paper, a CNN-based (Convolutional Neural Network-based) handwritten character recogni-
tion framework is proposed. In this framework, proper sample generation, training scheme and
CNN network structure are employed according to the properties of handwritten characters.
Here we discuss handwritten character recognition using a CNN model called AlexNet.
889

2 RELATED WORKS
In the area of Malayalam character classification, previous works have mainly focused on
traditional feature extraction methods, which take the most time since it contains the larger
number of character classes. Different proposed methods having high recognition rates were
reported for the handwritten recognition of Chinese, Japanese, Tamil, Bangla, Devanagari
and Telugu using CNN.
El-Sawy (2017) proposed handwritten Arabic character recognition using CNN. The CNN
model was trained and tested using a database of 16,800 handwritten Arabic character images
with an average misclassification error of 5.1% for the test data.
Tsai’s (2016) model is a Deep Convolutional Neural Network (D-CNN) for recognizing
handwritten Japanese characters. A VGG-16 network with 11 different convolutional neural
network architectures were explored. The general architecture consists of a relatively small
convolutional layer followed by an activation layer and a max pooling layer with a final FC
(Fully Connected) layer having the same number of channels as the number of classes. It
achieved an accuracy rate of 99.53% for overall classification.
Roy et al. (2017) introduced a layer-wise deep learning approach for isolated Bangla hand-
written compound characters. Supervised layer-wise trained D-CNNs are found to outper-
form standard shallow learning models such as Support Vector Machines (SVM) as well as
regular D-CNNs of similar architecture by achieving an error rate of 9.67%, and thereby
setting a new benchmark on the CMATERdb 3.1.3.3 with a recognition accuracy of 90.33%,
representing an improvement of nearly 10%.
For Chinese (Xuefeng Xiao et al., 2017) to incur the high computational cost of deeper
networks a Global Supervised Low-Rank Expansion (GSLRE) method and an Adaptive
Drop-Weight (ADW) technique was used. For HCCR with 3,755 classes a CNN network
with nine layers was adopted, that can reduce the networks computational cost by nine times
and compress the network to 1/18 of the original size of the baseline model, with only a
0.21% drop in accuracy.
Md. Mahbubar Rahman et al. (2015) considered a CNN-based Bangla handwritten
character recognition system. The normalized data was used and employed CNN to clas-
sify isolated character recognition. 20,000 handwritten characters with different shapes and
variations were used in this study. The proposed BHCR-CNN misclassified 351 cases out of
2,500 test cases and achieved an accuracy of 85.96%. On the other hand, the method misclas-
sified 954 characters out of 17,500 training characters giving an accuracy rate of 94.55%.
An integrated two classifier – CNN and SVM – model for Arabic character recognition
was introduced by Mohamed Elleuch and Kherallah (2016) with a dropout technique thereby
reducing the over fitting. The performance of the model was compared with the character
recognition accuracies gained from state-of-the-art Arabic Optical Character Recognition.
The error rate without dropout layer was recorded as 14.71% and a considerable reduction
error rate of 5.83% using dropout was reported.
An unsupervised CNN model for feature extraction and classification of multi script rec-
ognition was proposed by Durjoy Sen Maitra and Parui (2015). For a larger character class
problem, they performed a certain amount of training for a five-layer CNN. SVM was used
as the classifier for six different character databases all of which have achieved an error rate
less than 5%.
Another approach for handwritten digit recognition using CNN and SVM for feature
extraction and recognition respectively (Chunpeng Wu et al., 2014) achieved a recognition
rate of 99.81% without rejection, and a recognition rate of 94.40% with 5.60% rejection.
Jinfeng Bai et al. (2014) proposed a Shared Hidden Layer deep Convolutional Neural Net-
work (SHL-CNN), which recognizes both English and Chinese image characters, produced a
reduced recognition error of 16–30%, compared with models trained by characters of only one
language using conventional CNN, and by 35.7% compared with state-of-the-art methods.
Zhong et al., (2015) presented a deeper CNN architecture for handwritten Chinese
character recognition (denoted as HCCR-GoogLeNet) which uses 19 layers in total.
890

The work also uses certain directional features like HoG, Gabor filters and gradient features
to improve the performance of HCCR-GoogLeNet. It involves 7.26 million parameters, and
has shown that with the proper incorporation with traditional directional feature maps, the
proposed model achieved new state-of-the-art recognition accuracy of 96.35% and 96.74%,
respectively, outperforming the previous best result by a significant gap.
An offline English word recognition method using neural networks has been proposed by
Gupta et al. (2011). It divides the word recognition process into two classes, segmentation
based and holistic based. A second approach is used in identification of fixed size vocabulary
where global features extracted from the entire word image are considered. The segmentation
based (Bhanushali et al., 2013) strategies, on the other hand, employ bottom-up approaches,
starting from the stroke or the character level and going toward producing a meaningful word.
Another framework for handwritten English character recognition presented by Yuan
et al. (2012) use a modified LeNet-5 CNN model. It uses a different architecture of neurons
for each layer. An error-samples-based reinforcement learning strategy is developed for train-
ing of the CNN. UNIPEN lowercase and uppercase datasets were used for evaluating the
experiments, with recognition rates of 93.7% for uppercase and 90.2% for lowercase letters.
The majority of the published neural networks employed only fully connected layers, while
most of the efforts were focused on feature extraction.
3 PROPOSED METHOD
Handwritten Malayalam character recognition faces challenges such as similarity in structure

of characters, unlimited variations in human handwriting, un avail ability of standard datasets
and much more. Even though many handcrafted feature extraction methods were used for
Malayalam character recognition none has achieved 100% accuracy. Here we are proposing
an automated feature extraction technique using CNN that efficiently extracts features from
character images and produces higher recognition rates. This work also focuses on compound
Malayalam character recognition using CNN. The overall flow of data is shown in Figure 2.
The proposed method has the following stages:
3.1 Data collection

Malayalam character recognition has been a recent topic of research and it poses the limita-
tion of unavailability of standard datasets for classification. Our initial work is to collect the
required amount of handwritten character data from different people. Collecting data from
different people will help the system to learn variations and to predict exactly while testing.
A larger dataset results in more accurate prediction of the test images.
3.2 Data augmentation

Unavailability of standard datasets and limitations for collecting a large amount of hand-
written dataset causes problems when using CNN since it needs a huge amount of data for
training and testing. To transform and modify the available data is one method to increase the
size of dataset. Here we use affine transformation which is a linear transformation method
to augment the data. To preserve points, straight lines, and planes, a linear mapping method
is used called affine transformation. Translation, scaling, sheering and rotation are the four
major affine transformations.
3.3 Preprocessing
Performance of any character recognition system is directly dependent upon the quality of
the input documents. To remove the noise from the character images a data preprocessing
method is used. Salt and pepper noise are the most common noise elements present in an
891

image and these can be removed by some contrast enhancement processes. The input image
also requires some resizing. The image size should fit the network and also be recognizable.
3.4 Model training and optimization

The model used for training the data is the AlexNet model which is a basic framework of CNN.
The AlexNet architecture (shown in Figure 1) consists of eight learned layers—five convolu-
tional and three fully connected (Krizhevsky et al., 2012). Some of the most important layers
are ReLU that uses max(0,x) function which train several times faster than their equivalents
with tanh units. Also, they do not require input normalization to prevent them from saturating.
The next layer is max-pool layers, which summarize the outputs of neighboring groups of neu-
rons in the same kernel map. To reduce over fitting, we use dropout layers. To reduce the model
training error, we use a weight decay mechanism. To improve the training and testing accuracy
after each process, the parameters set has to be optimized with different values to check the
point which produces the maximum accurate predictions. Some of the parameters used here
are batch size, momentum, weight decay, learning rate, iteration index and epoch.
3.5 Testing
Testing module deals with the test images. Test images are obtained by splitting the augmented
dataset randomly. It will first preprocess the input image and it will classify the unlabeled test
Figure 1. AlexNet architecture.
Figure 2. Sketch of flow configuration.
892

data. Test data is not labeled in the sense it should be recognized by the machine. Labels are
assigned to each of the test images by the network and then the accuracy is measured. In the
post processing stage of the proposed system, the classifier output is mapped to the character
Unicode. The output of the classifier will be some integer labels. This integer label should be
converted into the corresponding character Unicode. The Unicode is written in a text file.
3.6 Classification
The final layer of the CNN is a softmax layer (Pranav P Nair et al., 2017) and this softmax
layer is used for classifying the given input image. This softmax layer is used to classify the
character. The softmax function has a value between 0 and 1. The sum of the output of
all the classes is also 1. The class with the maximum value will be selected as the class for a
particular input image.
4 CONCLUSIONS AND FUTURE WORK
Character recognition has a wide range of applications in postal automation, tax returns, bank
cheque processing and many more. Even though works on Malayalam character recognition
have been reported none has achieved 100% accuracy and most of them take a large processing
time.
Our proposed method uses automatic feature extraction using CNN that considerably
reduces the training and testing time and produces a more accurate result. Here compound
Malayalam character recognition is also considered, which makes the system more useful for
real time application processing. The model needs a system with Core i7 PC of 2.6 GHz with
64GB memory with CUDA enabled GPU for parallel processing.
REFERENCES
Bhanushali, S., Tadse, V. & A. Badhe, A. (2013). Offline handwritten character recognition using neural
network. In Proceedings of National Conference on New Horizons in IT-NCNHIT (pp. 155).
Chunpeng Wu, Wei Fan, Y.H.J.S.S.N. (2014). Handwritten character recognition by alternately trained
relaxation convolutional neural network. In 14th International Conference on Frontiers in Handwriting
Recognition (pp. 291–296).
Durjoy Sen Maitra, U.B. & Parui, S.K. (2015). CNN based common approach to handwritten character
recognition of multiple scripts. 13th International Conference on Document Analysis and Recognition
(ICDAR) (pp.1021–1025).
El-Sawy, A. (2017). Arabic handwritten characters recognition using convolutional neural network.
WSEAS Transactions on Computer Research 5, 2415–1521.
Gupta, A., Srivastava, M. & Mahanta, C. (2011). Offline handwritten character recognition using neural
network. In IEEE International Conference on Computer Applications and Industrial Electronics
(ICCAIE) (pp. 102–107). New York, NY: IEEE.
Jinfeng Bai, Zhineng Chen, B.F.B.X. (2014). Image character recognition using deep convolutional
neural network learned from different languages. In IEEE International Conference on Image
Processing (ICIP) (pp. 2560–2564).
Krizhevsky, A., Sutskever, I. & Hinton, G.E. (2012). Imagenet classification with deep convolutional
neural networks. In Pereira, F., Burges, C., Bottou, L. & Weinberger, K. (Eds). Advances in Neural
Information Processing Systems, 25, pp. 1097–1105. Curran Associates, Inc.
Md. Mahbubar Rahman, M.A.H. Akhand, S.I.P.C.S. (2015). Bangla handwritten character recognition
using convolutional neural network. International Journal of Image, Graphics and Signal Processing 8,
42–49.
Mohamed Elleuch, R.M. & Kherallah, M. (2016). A new design based-SVM of the CNN classifier
architecture with dropout for offline Arabic handwritten recognition. In International Conference on
Computational Science (ICCS) 80, (pp. 1712–1723).
Pranav P Nair, Ajay James, C.S. (2017). Malayalam handwritten character recognition using
convolutional neural network. International Conference on Inventive Communication and
Computational Technologies (ICICCT 2017) (pp. 278–281).
893

Saikat Roy, Nibaran Das, M.K.M.N. (2017). Handwritten isolated Bangla compound character
recognition: A new bench-mark using a novel deep learning approach. Elsevier Pattern Recognition
Letters, 90, 15–21.
Tsai, C. (2016). Recognizing handwritten Japanese characters using deep convolutional neural networks.
http://cs231n.stanford.edu/reports/2016/pdfs/262_Report.pdf.
Xuefeng Xiao, Lianwen Jina, Y.Y.W.Y.J.S.T.C. (2017). Building fast and compact convolutional neural
networks for offline handwritten Chinese character recognition. Elsevier Pattern Recognition 72,
72–81.
Yuan, A., Bai, G., Jiao, L. & Liu, Y. (2012). Offline handwritten English character recognition based on
convolutional neural network. In 10th IAPR International Workshop on Document Analysis Systems
(DAS) (pp. 125–129). New York, NY: IEEE.
Zhong, Z., Jin, L. & Xie, Z. (2015). High performance offline handwritten Chinese character recognition
using GoogLeNet and directional feature maps. In 13th International Conference on Document
Analysis and Recognition (ICDAR) (pp. 846–850). IEEE.
894

A novel technique for script identification in trilingual optical

character recognition
A. Rone Maria & K.J. Helen

ABSTRACT: In multilingual environment searching, editing and storing of documents

is made easier by script identification. It also an aid for selecting script specific Optical
Character Recognition (OCR) for multilingual documents. India is a multilingual country
so documents may contain more than one script. In Kerala, a state in India, the documents
may contain text in three languages: Malayalam, the—official language of the state; Hindi,
the national language; and English, the global language. For processing such multiscript
documents, it is necessary to identify the script before feeding the text line to specific OCRs.
This paper presents a novel and efficient technique for script identification in English, Hindi,
Malayalam trilingual documents. Features for classification are extracted from horizontal
projection of text images. Training and testing are done on our own data set developed from
documents containing these three languages.
1 INTRODUCTION
Optical Character Recognition (OCR) is the technology of translating text in an image to

machine encoded form, which has gained huge importance in the development of document
analysis systems for applications such as digitization, machine translation, and cross-lingual
information. The idea of an OCR system is to identify alphabets, numbers, punctuation
marks, or special characters, present in digital images, without any human involvement. This
is achieved through a matching process between the extracted features of a given character’s
image and the library of image models. Previously, most of the research work of OCR has
dealt with only monolingual documents, but actual documents may contain more than one
script. In multilingual imaged document analysis and processing there are many fields where
good research has been carried out. Some of the applications to recognize multilingual text
are for postal addresses, data entry forms, railway reservation forms, bank cheques, applica-
tion forms, question papers, government forms, receipts, magazines, and newspapers which
contain text written in two or more languages. In India every government office uses at least
three languages: English’ Hindi, the official language of the nation; and the official language
of the corresponding state (Padma et al., 2009). In Kerala, a state in India, the documents
may contain text in three languages; Malayalam, the official language of the state; Hindi,
the national language; and the English global language. Figure 1 shows some examples of
trilingual documents in Kerala. There are two distinct methods to build Multilingual Optical
Character Recognitions (MOCRs): i) a single OCR is trained for all the scripts, or ii) train
an isolated OCR for each script (Mathew et al., 2016). In the first method, the classifier
would need a huge collection of data and the output space is much larger. The training of the
classifier becomes time-consuming and computationally expensive with larger output space.
But, in the second method, the data required for training would be much lesser and output
space would be smaller. Most of the work done in recognizing multilingual scripts has been
restricted to using a script identification stage before recognition. This is due to the fact that
either the segmentation or the recognition relied on script-dependent features. Other groups
895

Figure 1. Illustration of some examples of multiscript documents in English, Hindi and Malayalam.
use combined training data for both scripts, but such OCRs have not been very successful
due to the huge search required in a large database. Such OCRs also suffer from errors when
words are classified as belonging to the wrong script (Kaur & Mahajan, 2015). Thus, most
MOCRs perform word segmentation followed by script identification and then recognition.
This paper discusses the script identification for English, Hindi and Malayalam scripts in
documents. This is achieved by extracting features from the horizontal projection profile of
text lines.
The rest of the paper deals with some of the previous works in section 2 and proposed
work in section 3. This paper concludes in section 4.
2 PREVIOUS WORKS
Philip and Samuel (2009) used dominant singular values and Gabor features for classification of
printed English and Malayalam script. To identify the script the text is segmented to word level
and Gabor features are extracted. Dominant singular values are used for character recognition.
896

Rahiman et al. (2011) proposed another method for an English-Malayalam bilingual OCR.
Here the image is scanned from top left to bottom right and segmented into line level, word
and character level. This segmented character is resized to a 16 X 16 bitmap. A database is
created for characters in different fonts for both languages. The resized image is compared
with characters in the database. A pixel match algorithm is used for this comparison and the
matched character is displayed.
Aithal et al. (2010) also proposed a simple and efficient method for script identification
from a trilingual document in English, Hindi and Kannada languages. The proposed system
uses a horizontal projection profile for script identification.
Mohanty et al. (2009) have developed a bilingual OCR for recognizing English and Oriya
texts. Structural features such as the presence of loops, concavities, lines present or absent
on the left and right part of characters, upper portion open are selected as features in this
work.
A method for identifying languages from documents printed in three languages English,
Hindi, Kannada has been proposed by Padma et al. (2009). The proposed approach is based
on the characteristic features of the top and bottom profiles of input text lines.
Kaur and Mahajan (2015) have proposed an approach for the identification of English
and Punjabi scripts at line level. Character density features and headlines are used for script
identification.
Dhandra et al. (2007) proposed a method for script identification at word level. They deal
with scripts such as Kannada, Tamil, Devnagari and English numerals using visual appear-
ance. Seven features are extracted in real values and form a vector using these for each train-
ing word. K nearest neighbor classifier is used for classifying test words.
3 PROPOSED WORK
This paper presents a novel script identification technique for identifying English, Hindi and
Malayalam scripts from multiscript documents. The steps of the proposed system are shown
in Figure 2.
897

3.1 Image acquisition
The input is the image of the trilingual document. Converting the paper document into an
image by scanning is the first step. A scanned imaged can be in any format, jpg, jpeg or png.
The scanned image is transformed to a gray scale image.
3.2 Preprocessing
Image preprocessing is the process of improving image quality for better understanding using
predefined methods. Commonly used preprocessing methods are noise removal, binariza-
tion, and skew correction. The noise may occur during scanning or transferring of document
images. Smoothing operations are used for noise removal. Converting gray scale image from
256 gray levels to two levels is called binarization. The binarization is generally done by tak-
ing a threshold value for an image and set intensity values to one for pixels which have larger
value than threshold value. Set intensities to zero if it is less than the threshold value. Skew is
a deformation that is introduced while scanning a document. It is necessary for aligning the
text lines to the coordinate axes.
3.3 Segmentation
White gap between text lines are used for segmentation. Horizontal projection of the scanned
image is used for line segmentation. Horizontal projection of a trilingual document is shown
in Figure 3. The zero value points in the figure represents the spaces between text lines. Line
segmentation is performed at these points.
3.4 Feature extraction

Feature extraction is an important phase of any character recognition system. The objective
of feature extraction is to ensure pattern identification with a minimum number of features.
In this proposed system, the features are extracted from the horizontal projection profile. The
horizontal projection profile is obtained by counting the number of black pixels in each row
of the image (Puri and Singh, 2016). Figure 4 shows feature extraction from the horizontal
projection of text lines. From this projection, the peak points P1 and P2 are determined. The
values at peak points P1 and P2 are the features used for classifying the English, Hindi and
Malayalam scripts.
Figure 3. Horizontal projection of a trilingual document.
898

Figure 4. Feature extraction from horizontal projection.
Table 1. Range of P value for

different scripts.
Range of
Script P value
English 1.10 to 1.30

Malayalam 0.85 to 0.95
Hindi 2.00 to 2.50
3.5 Classification
Classification is based on the features extracted in the previous step. For this proposed sys-
tem, a rule-based classifier is used.
STEP 1: For each text line determine the first and second peak points P1 and P2 respectively.
STEP 2: Find the values at P1, value (P1), and P2, value (P2).
STEP 3: Calculate the decision parameter Pvalue from these values as
value (P1)
P value = (1)
value (P 2 )
STEP 4:
If P value falls in the range of 0.85 to 0.95 then the text is identified as Malayalam.
If P value falls in the range of 1.10 to 1.30 then the text is identified as English.
If P value falls in the range of 2.00 to 2.50 then the text is identified as Hindi.
The range of Pvalues for English, Hindi and Malayalam scripts are shown in Table 1.
Based on these observed values, rules are made and fed into the rule-based classifier.
4 CONCLUSIONS
In this paper, a novel technique for script identification in English, Hindi and Malayalam
multilingual documents has been proposed. Features for classification are extracted from the
horizontal projection of scripts. A rule-based classifier is built from the knowledge obtained
from the sample data.
REFERENCES
Aithal, P.K., Rajesh, G., Acharya, D.U. & Subbareddy, N.K.M. (2010). Text line script identification
for a tri-lingual document. In International Conference on Computing Communication and Networking
Technologies (ICCCNT) (pp. 1–3). New York, NY: IEEE.
899

Dhandra, B., Hangarge, M., Hegadi, R. & Malemath, V. (2007). Word level script identification in bilin-
gual documents through discriminating features. In International Conference on Signal Processing,
Communications and Networking (ICSCN’07) (pp. 630–635). New York, NY: IEEE.
Kaur, I. & Mahajan, S. (2015). Bilingual script identification of printed text image. International Journal
of Engineering and Technology 2(3), 768–773.
Mathew, M., Singh, A.K. & C. Jawahar, C. (2016). Multilingual OCR for Indic scripts. In 12th IAPR
Workshop on Document Analysis Systems (DAS) (pp. 186–191). New York, NY: IEEE.
Mohanty, S., Dasbebartta, H.N. & Behera, T.K. (2009). An efficient bilingual optical character
recognition (English-Oriya) system for printed documents. In Seventh International Conference on
Advances in Pattern Recognition (ICAPR’09) (pp. 398–401). New York, NY: IEEE.
Padma, M., Vijaya, P. & Nagabhushan, P. (2009). Language identification from an Indian multilingual
document using profile features. In International Conference on Computer and Automation Engineering
(ICCAE’09). (pp. 332–335). New York, NY: IEEE.
Philip, B. & Samuel, R.S. (2009). A novel bilingual OCR for printed Malayalam-English text based on
Gabor features and dominant singular values. In International Conference on Digital Image Processing
Puri, S. & Singh, S.P. (2016). Text recognition in bilingual machine printed image documents—challenges
and survey: A review on principal and crucial concerns of text extraction in bilingual printed images.
In 10th International Conference on Intelligent Systems and Control (ISCO) (pp. 1–8). New York,
NY: IEEE.
Rahiman, M.A., Adheena, C., Anitha, R., Deepa, N., Kumar, G.M. & Rajasree, M. (2011). Bilingual
OCR system for printed documents in Malayalam and English. In 3rd International Conference on
Electronics Computer Technology (ICECT), 3 (pp. 40–45). New York, NY: IEEE.
900

An optimal controller placement strategy using exemplar-based

clustering approach for software defined networking
A.G. Sreejish & C.N. Sminesh

Department of Computer Science and Engineering, Government Engineering College, Thrissur, Kerala
ABSTRACT: Software Defined Networking (SDN) is a next-generation paradigm that

emphasizes the decoupling of networking control plane and the data plane. The decoupling
allows managing the network using software programs in a flexible and efficient manner.
The feasibility, scalability, and performance of large-scale SDN are confronted with critical
challenges, need multiple controllers to impact performance outlook of SDN. The controller
placement and load balancing in SDN are critical problems in providing a reliable network
with high resource utilization. To resolve these problems, an effective controller placement
strategy is needed. This paper focuses on minimizing latency between controllers and switches
and maximizing load balance ability of the controllers. In order to obtain the optimal number
of controllers and its corresponding locations, a modified affinity propagation algorithm is
proposed which takes into consideration the computation complexity of the actual network
state. The proposed mechanism provides a solution by considering the trade-offs between
latency and load-balance.
1 INTRODUCTION
The emerging technology Software Defined Networking (SDN) emphasizes the decoupling
of control plane and the data plane. This separation helps to provide only an abstract view of
network resources and its state to the external applications. In large-scale SDN, the control
plane is composed of multiple controllers that provide the global view of the entire net-
work. The controllers are programmable, which acts as an intermediary between the network
administrator and data plane. In control plane, control intelligence is decoupled. Therefore,
The controllers install the rules in the flow table that are used to forward the traffic flows
entering the network and switching is done by OpenFlow (OF) switches in the forwarding
plane. However, the separation of control-data plane introduces performance limitations and
reliability issues (Yeganeh et al. 2013, Jarschel et al. 2012). They are:
a. The nodes in the network must be continuously monitored and controlled by using a
proactive or reactive method. The nodes communicate with the corresponding controllers
to obtain the new forwarding rules to be installed. Based on these rules, the nodes process
various new flows arriving at it. The response time of the overall system will increase when
the communication overhead between controllers and switches is high. This is because the
controller has a limited processing power (Yeganeh et al. 2013), with respect to the number
of nodes assigned to it or the number of flow queries is too high.
b. In large-scale SDN, the density of network elements and traffic flows are really high. If
we are using a single physically centralized controller, there is a chance for Single Point
of Failure (SPoF). So multiple controllers have to be placed to ensure that SPoF will
be eliminated from the network. In SDN, control plane creates the logically centralized
view of the entire network. Inorder to create this view, the controllers need to communi-
cate with each other and update/synchronize their databases (Tootoonchian and Ganjali
2010). To reduce the inter-communication among the controllers, we can create an overlay
network linking the controllers (Shi-duan 2012).
901

c. The failure of controller results in disconnection of data plane and control plane easily
may lead to packet loss and performance degradation (Yeganeh et al. 2013, Heller et al.
2010). Consider the scenario that some of the nodes are still alive, if they are unable to
communicate with the controllers, they may fail to process the newly arrived flows.
The controllers can be placed optimally by finding the solution for (a), (b), and (c), which
will be an NP-hard problem (Heller et al. 2010). As a result, various solutions have been
suggested focusing on performance, (problem (a), (b)), and performance plus reliability
(problem (c)).
In this paper, we analyze some of the solutions and then proposes a controller placement
strategy that improves availability of the network by considering both performance and reli-
ability. The controller placement strategy affects all aspects of performance of SDN like the
delay between a controller and a node, network availability, and other performance metrics.
So the aim is to develop an algorithm that computes the number of controllers needed and
the location of controllers for a large-scale network where the placement results in significant
improvement of both performance and reliability of the network. To reduce the computation
cost of finding the optimal placement of controllers an exemplar-based clustering technique
called affinity propagation is used.
In this paper, the remaining sections are organized as follows. Section 2 presents the related
works. Section 3 provides an overview of the proposed system and several metrics that are
used to formulate the final algorithm. Section 4 comprises of conclusion and future work.
2 RELATED WORKS
This section gives an overview of some related works on controller placement in large-
scale SDN. The main problem is to find the number of SDN controllers that are required
for a given network topology and where to place them so that it resolutely maintains the
performance even in cases of failure
In WANs, if the placement of controller introduces a significant increase in path delays,
then it will postpone the time taken for control plane to achieve the steady state. It has been
explained in (Heller et al. 2010) which is theoretically not new. If only propagation latency is
considered, the issue is akin to a warehouse or facility location problem, which can be solved
by the use of Mixed Integer Linear Program (MILP) tools.
The Heller et al. prior work (Heller et al. 2010) encourages to consider the problem of
controller placement and measures the impact of controller placement on existing topologies
like Internet2 (Yeganeh et al. 2013) and on various cases available in the Internet Topology
Zoo (Knight et al. 2011). In reality, the main purpose was not to locate the ideal positions
for the controllers but to provide an evaluation of a major design problem which requires
further examination. It has been demonstrated that optimal solutions can be discovered for
practical network instances in failure-free cases by figuring out the complete solution with
offline calculations. This work also affirms that in most topologies the existing response time
requirements cannot be satisfied using a single controller, although they didn’t consider the
resiliency aspects.
(Zhang et al. 2016) address Multi-objective Optimization Controller Placement (MOCP)
problem and focus on three objectives; maximize controller load balance capacity, maxi-
mize network reliability and minimize control path latency. It provides an optimal controller
placement strategy in such a way that the routing requests are optimally distributed among
multiple controllers. The work converted the MOCP into a mathematical model as the
optimization objective function and developed Adaptive Bacterial Foraging Optimization
(ABFO) algorithm to resolve it, claiming those above objectives are optimized efficiently and
effectively. But in this method, the optimal number of controllers needed not is identified
dynamically. For a large-scale network, this identification becomes exhaustive and will reduce
the reliability of the solution.
The work (Borcoci et al. 2015) presented an analytical view on using multi-criteria deci-
sion algorithms (MCDA) to choose an optimal solution from several controller placements
902

solutions in large-scale SDN with the help of some weighted variables. The advantage of
MCDA is that it can produce an optimal result at the same time as considering numerous
criteria. This method claims that it can be applied to different scenarios including reliabil-
ity-aware or failure-free assumption ones, primarily based on more than one metrics sup-
ported by using the reference model MCDA. This work deals with several metrics such as
average-case latency or worst-case latency in failure-free situations, worst-case latency when
a controller fails, nodes or links failures, inter-controller latency, controller load imbalance
and multi-path connectivity metric, but some of these metrics are considered only when the
service provider policy demands it. By using this method different network/service providers
gain the ability to maintain the quality of service (QoS) provided by them. This is achieved
by assigning weights to criteria when selecting the optimal solution. The limitation of this
work is that it demands to pre-compute all possible solutions for the given network topology,
thereby increasing the computational cost.
In order to reduce the computational complexity, cluster-based approaches can be used. For
dividing a large-scale SDN into small network domains, an efficient Spectral Clustering place-
ment algorithm is used (Xiao et al. 2014). With the aid of dividing WAN into various SDN
domains, the controller with fewer nodes can gain maximum performance. Also, there will be
a sharp reduction in the inter-controller communications. For examining metrics of controller
placement problem for SDN, they have considered the propagation latency and controller per-
formance and used Internet2 topology to probe the performance of the system. The number of
controllers as input for partitioning the network was the limitation of this work.
In (Zhao et al. 2016), a multi-controller placement problem that considers minimization
of the latency between controllers and switches is discussed. A modified version of affinity
propagation (AP) is used to resolve the controller placement problem. Affinity propagation is
an exemplar-based clustering method. The advantage of this method is that it can adaptively
compute the most favorable number and placement of controllers based on the network
topology. When comparing latency minimization with k-median and k-center methods, it
has size-able simulation results. This implies that this technique can offer more steady and
precise outcomes.
But, the propagation latency considered in above work is calculated only by considering
the shortest path between nodes. The latency can be affected by both the distance between
nodes and the bandwidth of the corresponding link. The objective of the proposed system
is to formulate a new model that takes both distance and bandwidth into consideration for
computation of similarity values used in affinity propagation.
3 PROPOSED SYSTEM
The main goal of the proposed system is to design an approach that automatically computes
the optimal number of controllers that are needed to manage the given SDN network and its
corresponding locations.
3.1 Proposed system design

This section outlines the overview of the architecture and some useful metrics that can be
used to evaluate the proposed system. The ultimate challenge is to maximize the Control
Plane performance. Among several metrics considered in (Borcoci et al. 2015), latency and
load of the controller are the major factors that affect the overall performance of SDN. For
the reason that the controller is responsible for producing forwarding rules, and the mis-
matched packets are buffered or discarded until the corresponding flow entries are installed.
The minimization of average-case latency can be considered as a k-median problem and
minimization of worst-case latency as k-center problem. However, to solve both k-median
and k-center problem, two input parameters; the number of controllers which is denoted by
k and the set of initial cluster centers which is denoted by C are needed. Another challenge
is controllers load balancing. For identifying the maximum possible load a controller can
903

Figure 1. Overview of proposed system.
process, the bandwidth of links connected to it can be considered. The overall design of the
proposed method is given in Figure 1.
The proposed system is advanced in four submodules; network topology creation, link
weight calculation, network partition and controller placement. In the first module, network
topologies are created using GARR and GEANT (Knight et al. 2011). The bandwidth of the
links present in these topologies and latency are considered to derive link weight. The third
module partition the network using modified affinity propagation. Finally, the controllers
are placed in the locations of exemplars.
For the modified-Affinity Propagation method, there is no need to initialize the number
of controllers and its locations. Affinity Propagation (AP) is an exemplar-based clustering
approach that has numerous benefits which include performance, no need to initialize the
value of k, and ability to obtain exemplars with high accuracy. The AP is modified in such a
way that it adapts to the problem of controller placement in SDN. Especially, the similarity
measurement between two nodes adopts both latency and bandwidth of the links.
In a real network topology, there may be no direct links between two nodes. But they are
reachable via other links present in the network. So the reachable shortest path distance
L(u, v) is considered as latency (Zhao et al. 2016), which can be found by using the Floyd-
Warshall all-pairs shortest paths algorithm (Cormen et al. 2009). The bandwidth of the links
is already available in the internet topologies GARR and GEANT.
The exemplar-based clustering problem is formulated by using the link weight which is
assigned to each edge, considering the fact that it minimizes the latency and equalizes the
load of the controller. Link weight is composite metric, which is computed using Eq. 1.
An assumption is made that the locations of controllers are the same as some of the nodes.
LinkWeight (u, v) = I(u, v)(BW(u, v) + L(u, v)) (1)

where I(u,v) is used to indicate the presence of a link between u and v. BW(u, v) is the band-
width of the link present in between u and v and L(u, v) is the latency.
904

3.2 Performance metrics
Measuring the performance of the proposed technique is vital to evaluating how well it solves
the controller placement problem. The metrics that can be used to evaluate the proposed
method are average and worst case latency, controller imbalance factor and inter-controller
latency which is discussed in the following section.
3.2.1 Average and worst case latency

The network can be represented by using an undirected graph G(V, E) where V is the set of
nodes and E is the set of links present in it. n denotes the number of nodes and it is defined
as n = |V|.
Latency is a measure of how much time it takes for a packet of data to reach from one
node to some other node, which can also be computed in terms of distance. Here, the latency
is considered as the shortest path distance between two nodes u and v. There are two vari-
ations of latencies that can be defined for a particular controller placement; average case
latency and worst case latency (Borcoci et al. 2015).
Average Case Latency:
1
Lavg = ∑ min L(v, c)
n v ∈V c ∈C
(2)
Worst Case Latency:
Lworst = max min L(v,c) (3)

v ∈V c ∈C
Any of above mentioned latencies need to be minimum for the optimization algorithm that
should obtain a placement of controller.
3.2.2 Controller imbalance factor

A properly designed system would require more or less same load on all controllers, i.e.,
consistency of the node to controller distribution. A metric can be introduced to measure the
imbalance between controllers for a given controller placement. It is defined as the difference
between the most and minimal number of nodes assigned to a controller (Zhao et al. 2016).
Cimbalance = max nc − min nc (4)

c ∈C c ∈C
where nc denotes the number of nodes under the controller c. For Eq. 4, need to take
consideration that during the case of failures the control of nodes may be moved from
primary controller to other controllers. This reassignment can increase the load of respective
controllers. An optimization algorithm should minimize Eq. 4 in order to find a controller
placement that provides good performance.
3.2.3 Inter-controller latency

The inter-controller latency (Borcoci et al. 2015) will affect the response time of the inter-
controller communications, which are performed to update the global view of the network status.
For a given placement, it can be computed as the maximum latency between two controllers:
Lint er − controller = max L( cu , cv ) (5)
where L(cu, cv) is the distance between two controllers cu and cv. For minimizing Eq. 5, we
need to place the controllers close to each other. But the problem is that, this may increase
the node-controller latencies given by Eq. 2 and Eq. 3.
905

The modified-AP is implemented using python and above performance metrics are used
to evaluate the proposed system using Mininet on two topologies GARR and GEANT.
GARR topology is composed of 52 nodes and 66 edges, and there are 40 nodes and 61 edges
present in GEANT topology. We can compare the results of the modified-AP algorithm with
k-median and k-center to verify the stability and accuracy of the proposed system.
4 CONCLUSION AND FUTURE WORK
This paper has shown a work (in progress) study on the application of several weighted criteria
for placement of controller in large-scale SDN. The proposed method dynamically compute
the number of controllers needed and its location for a given network using modified affinity
propagation (modified-AP) clustering. This approach additionally specifies the most suitable
controller for each switch. This method can be applied to a scenario that considers failure-
free assumptions, given that it attains an overall optimization.
As future work, the proposed method can be extended to include fault-tolerance and reli-
ability aspects to improve the efficiency of the controller placement. The simulations can
be conducted on large-scale SDN by considering additional metric like the capacity of the
controllers.
REFERENCES
Borcoci, E., R. Badea, S.G. Obreja, & M. Vochin (2015). On multi-controller placement optimization
in software defined networking-based wans. ICN 2015: The Fourteenth International Conference on
Networks, 261–266.
Cormen, T., C. Leiserson, & R. Rivest (2009). Introduction to algorithms. Massachusetts, USA: The
MIT Press.
Heller, B., R. Sherwood, & N. McKeown (2010). The controller placement problem. In Proc. HotSDN,
pp. 7–12.
Jarschel, M., F. Lehrieder, Z. Magyari, & R. Pries (October 2012). A flexible openflow-controller
benchmark. In Proc. European Workshop on Software Defined Networks (EWSDN), Darmstadt,
Germany, pp. 48–53.
Knight, S., H.X. Nguyen, N. Falkner, R. Bowden, & M. Roughan (2011). The internet topology zoo.
IEEE JSAC 29, 1765–1775.
Shi-duan (October 2012). On the placement of controllers in software defined networks. ELSEVIER,
Science Direct 19, 92–97.
Tootoonchian, A. & Y. Ganjali (2010). Hyperflow: a distributed control plane for openflow. In Proc.
INM/WREN.
Xiao, P.,W. Qu, H. Qi, Z. Li, & Y. Xu (2014). The sdn controller placement problem for wan. In
Communications in China (ICCC), 2014 IEEE/CIC International Conference on, pp. 220–224. IEEE.
Yeganeh, S.H., A. Tootoonchian, & Y. Ganjali (February 2013). On scalability of software-defined
networking. IEEE Comm. Magazine 51, 16–141.
Zhang, B., X. Wang, L. Ma, & M. Huang (2016). Optimal controller placement problem in internet-
oriented software defined network. In Proc. International Conference on Cyber-Enabled Distributed
Computing and Knowledge Discovery (CyberC), pp. 481–488.
Zhao, J., H. Qu, J. Zhao, Z. Luan, & Y. Guo. (2016). Towards controller placement problem for software-
defined network using affinity propagation. Electronics Letters 53, 928–929.
906

Identifying peer groups in a locality based on Twitter analysis
K. Sreeshma & K.P. Swaraj

Department of Computer Science Government Engineering College, Thrissur, Kerala, India
ABSTRACT: Twitter is one platform where people express their thoughts on any trending
topics they are interested in. The exploration of this data can help us to find peer groups or
group of users with similar interests. As in any other social network, this is also subjected to
various spam attacks. So before identifying peer groups, the accounts that are ingenuine or
regularly involved in spamming activities has to be filtered out. The main idea is to make use
of the URLs the accounts share and their frequency to identify the account type.Here instead
of focusing on one account, a group of accounts or a campaign is identified based on the
similarity of the accounts. The similarity measure is calculated by applying Shannon’s Infor-
mation theory to estimate the amount of information in a URL and then using the value to
find out information shared by each account. Once similar accounts are identified a graph
is plotted connecting those accounts who have a similarity measure above a threshold. The
potential campaigns are identified from this graph. Then they are classified to spammers and
normal users using ML algorithms. The normal users we thus identify are members who have
similar interests. To further improve the efficiency these members are grouped together based
on their location, so peer groups in a locality are identified. This peer group identification
can help in connecting those people with similar interests in a locality.
1 INTRODUCTION
The exploding growth of data has always opened new opportunities for those who indulge.
The role of social networks in the life of a person is unfathomable these days. In fact, the
scenario is like ones virtual friend knows about them more than their own friends and fam-
ily. So the behavioral data of an individual is all available in his social network accounts.
The right assessment of his social media activity can be helpful to identify his character
and interests.
Twitter is a microblogging social network where most of the users express their opinion on
some trending topics. The tweets they post or share will help us find the interest of the users.
The idea here is to identify a group a people with similar interest. The tweets generally allow
text and URLs only. So the URLs play a significant role in the characteristics of a tweet
(Zhang et al. 2016). To be able to identify the common URLs shared between the users will
be an important measure in establishing a relation between similar users.
Being a very popular social media network it has its disadvantages as well. It is also sub-
jected to spam attacks. The spamming in social networks can cause much more harm than
traditional spamming like email spamming (Grier et al. 2010). There will be a lot of ingenu-
ine users with fake profiles who like to promote illegitimate contents. So when we are to peer
groups based on their social network activity we should be able to genuine accounts from
spammers efficiently. After filtering out the spammers we need to rightly group the authentic
users with similar interests and then we can apply geotagging techniques to group them based
on their geographic coordinates.
In this article, we are here to identify peer groups in a locality based on their Twitter usage.
For this purpose, we are making use of the URLs that are part of their tweets. So as the first
step we need to connect the users which share similar interests based on an account similarity
907

measure derived on the basis of the number of common URLs they share and the informa-
tion contained in each URL. Once we obtain an account similarity graph, next step is to
classify potential campaigns to normal and abnormal and finally we group accounts in each
campaign according to their locations so as to obtain peer groups in one locality. The rest of
the paper carries the details for each of the aforementioned stages.
2 RELATED WORK
Much related works have been done in identifying spammers in a social network. Since most
of the efforts lie in, correctly classifying normal and abnormal groups based on their activi-
ties, here importance is given to rightly classify spammers and non spammers. Most tradi-
tional methods try to detect the spammers based on their messages sent or the activities of
their account. The message level detection (Benevenuto et al. 2009) checks each tweet posted
for any discrepancies or spam contents in any URLs mentioned. But this may require a real-
time processing as umpteen tweets are getting posted every hour.
The account level detection methods (Lee et al. 2010) examine the activity of the user
accounts, such as whether they have promoted spam contents, to find the authenticity of an
account and thus identify if it’s a spammer or not. But both of these methods have left many
spams unidentified at the end of the evaluation. Instead of classifying individual messages
and accounts, some papers have proposed identifying spam campaigns. A campaign refers to
a group of accounts who purposefully work towards the same goal. Spam campaigns often
contain accounts which post harmful information such as malware, virus etc.
(Hatanaka & Hisamatsu 2010) proposed a method to group users into distinct blacklist
groups based on the degree of similarity in their bookmarks and they reduced the rank of the
bookmarks promoted by this blacklists. A detection framework for spam campaigns based
on the similarity of the URLs shared by the accounts is proposed by (Gao et al. 2010).
This framework quantifies the similarity measure between accounts based on the URLs they
share, to draw a similarity graph and also put forward some characters of spam campaigns.
(Lee et al. 2010), (Lee et al. 2013) has proposed a content-driven approach for identifying
spam campaigns and categorizing it. This method employs a strategy to group users based
on text similarity but classification is done by manual inspection.
(Zhang et al. 2016), (Zhang et al. 2012) proposes a multilevel classification method for
identifying spammers and non-spammers where the first level includes classification into
normal and abnormal campaigns and further classification to identify spam and promotion
campaigns. This method makes use of the similarity measure between the users to plot a
similarity graph. The similarity measure was calculated based on the common URLs shared
by the accounts and later they extended the work to consider the timestamp as well. (Jiang
et al. 2016) identify the importance of finding peer groups in a social network for a friend
recommendation system.
3 PROPOSED METHOD
In this section, a new method is proposed for identifying peer groups in a locality based
on the Twitter activity of the users. First, The account similarity is estimated based on the
common URLs shared by them. A similarity graph is plotted based on these measurements
and then we have to extract potential campaigns from it. Secondly, for classification of this
campaigns into spammers and non spammers, we have to apply machine learning techniques
and classify them into normal and abnormal campaigns. In the final step, now that we have
obtained normal campaigns and filtered the spam campaigns, we have to make use of the
geotags to group the users in a normal campaign according to their locality. Thus we will
obtain peer groups in a locality.
The process flow is depicted in 1.
908

3.1 Estimation of account similarity
For estimating the account similarity, we make use of the URLs shared in the tweets. Those
accounts who share same URLs tends to be a part of the same campaign. For quantifying
the information contained in a URL, we have to make use of Shannon’s Information theory.
The information contained in a URL u, is defined as:
I(u) = −logP (u) (1)
and P(u) is the probability of u and dened as
#u
P(u ) = (2)
N
where # u is the number of tweets containing URL u in the corpus and N is the number of
all tweets containing URL(s).
In order to calculate the amount of information contained in all URLs shared posted by
account ai, we make use of the below formula,
I a (i ) = ∑ Num (u ) * I (u )
u ∈U i
i (3)
where Numi (u) is the number of tweets containing URL u posted by account ai. The amount
of information shared by accounts ai and aj through the sharing of common URLs is
collectively summated as
I a (ij ) = ∑
u ∈U i ∩U j
( Numi (u ) + Num j (u )) * I (u ) (4)
Finally, to estimate the similarities between accounts, we use:
I a (ij )
Sij = (5)
I a (i ) + I a ( j )
where 0 ≤ Sij ≤ 1
Now that we have obtained the measure of the similarity between various accounts, we
need to plot a graph combining the accounts which has a similarity measure above a particu-
lar threshold. The obtained graph is used to identify potential campaigns. The campaigns are
those areas in the graph which are very dense.We identify them using the concept of maximal
Figure 1. Process flow diagram.
909

coclique. Maximal coclique is defined as (vi,vj) for all neighbors, which represents the size of
the largest clique containing both vertexes vi and vj, and vj is a neighbor vertex of vi.
3.2 Classification of potential campaigns

The potential campaigns have been identified from the similarity graph. Next is another
important step which classifies each of these into normal or spam campaigns. Here we are
listing some of the features which will help us to classify the campaigns. These features are
used to train some classifier based on ML algorithms and used for prediction of the type of
campaign. Those features include:
Average Posting Interval (API): It is the average time difference between posting tweets
with URLs. The spammers tends to have a low time difference between posting tweets.
Average Number of tweets containing URLs (TUNum): This will give the count of the no.
of tweets posted by the account that is having URLs in them. It tends to be higher for spam
accounts compared to normal accounts.
Posting frequency of tweets with URLs (TUFrequency): This measure is similar to the
previous one. But this gives a weightage instead of giving an exact number. This value is also
higher for spam accounts.
Average of distinct URLs posted by each account (URLNum): Spammers will post URLs
from various domains and hence will have higher URL-Num value.
VURatio: This is the ratio of valid URLs out of the total URLs shared by the account.
The number of invalid or broken URLs shared by spammers tends to be higher.
Blacklisted URLs: The spam accounts might share URLs which are blacklisted in black-
lists such as Google Sage Browsing, PhishingTank etc. Any account with intentions of mis-
guiding a user will make use of these URLs.
3.3 Identifying peer groups based on the locality

Now we have classified the potential campaigns and identified the normal campaigns. This
is a group of people with similar interests which is evident from the number of the common
URLs posted by them. This group is subjected to further refining by grouping them based
on their locality. The locality of the account is available in the tweet data. Based on this we
can finally identify peer groups of a particular locality. This can be made use of by various
recommendation systems which would like to recommend products or plans on a location
specific basis.
4 CONCLUSION
The peer group identification based on Twitter is proposed and the methods involved for
each stages are detailed. We have used URL based estimations for finding similar accounts
and created a graph. The cohesive campaigns are then extracted. The extracted campaigns
are classified using some very important features and categorized as normal and spam cam-
paigns. The accounts in each normal campaigns are further categorized on the basis of their
location and we obtain peer groups in a particular location.
The twitter analysis throws insight to the nature of the twitter user and can help him to
connect with people having similar thoughts. This can be helpful to various recommendation
systems that work on a location specific basis.
REFERENCES
Benevenuto, F., T. Rodrigues, V. Almeida, J. Almeida, & M. Goncalves (2009). Detecting spammers and
content promoters in online video social networks. In Proceedings of the 32nd International ACM
Conference on Research and Development in Information Retrieval(SIGIR09) ACM, 620–627.
910

Gao, H., J. Hu, C. Wilson, Z. Li, Y. Chen, & B. Zhao (2010). Detecting and characterizing social spam
campaigns. In Proceedings of the 10th Annual Conference on Internet Measurement(IMC10). ACM,
35–47.
Grier, C., K. Thomas, V. Paxson, & M. Zhang (2010). Spam: The underground on 140 characters or less.
In Proceedings of the 17th ACM Conference on Computer and Communications Security (CCS10)
ACM, 27–37.
Hatanaka, T. & H. Hisamatsu (2010). Method for countering social bookmarking pollution using user
similarities. In Proceedings of the 2nd Conference on Networked Digital Technologies (NDT) Springer,
523–528.
Jiang, F., C.K. Leung, & A.G. M Pazdor (2016). Big data mining of social networks for friend rec-
ommendation. IEEE/ACM International Conference on Advances in Social Networks Analysis and
Mining (ASONAM) IEEE, 3 pages.
Lee, K., J. Caverlee, Z. Cheng, & D. Z Sui (2013). Campaign extraction from social media. ACM
Transactions on Intelligent Systems and Technology 5.
Lee, K., J. Caverlee, & S. Webb (2010). Uncovering social spammers: Social honeypots + machine learn-
ing. In Proceedings of the 33rd International ACM Conference on Research and Development in Infor-
mation Retrieval ACM, 435–442.
Zhang, X., Z. Li, S. Zhu, & W. Liang (2012). Detecting spam and promoting campaigns in the twitter
social network. In Proceedings of the IEEE International Conference on Data Mining (ICDM12).
IEEE, 1194–1199.
Zhang, X., Z. Li, S. Zhu, & W. Liang (2016). Detecting spam and promoting campaigns in twitter. ACM
Transaction. Web 10, 28 pages.
911

Route choice analysis of metros using smart card data
Sreemol Sujix
Computer Science and Engineering, Thejus, India Engineering College, Thrissur, Kerala, India
ABSTRACT: Route choice behavior is an interesting topic in the transportation area.

Metro systems have become a great solution for meeting the emerging traffic needs of large
cities. The arrival of metros indeed indicates the development strategy of a country. Given
the ‘origin to destination’ pairing of a metro trip, there are often multiple routes between
such pairings. Different routes are chosen by passengers with different preferences, and such
preferences have always been a point of interest in the transportation area. The data in this
study has been collected from passenger smart cards. This analysis will help operators to
improve their passenger service, and will also help passengers to plan their journeys.
Keywords: Route choice analysis, Smart card data, Data mining, Big data
1 INTRODUCTION
Metros have become the most demanded mode of transport for passengers due to their
speed, efficiency, time management, comfort, capacity to accommodate more passengers,
and so forth. It has become a necessary piece of infrastructure for a growing metropolitan
city. The use of metros has not only helped in decreasing road traffic but has also paved the
way to pollution-free transport when compared to cars and other vehicles that pollute the air
by emitting harmful carbon monoxide, which can create holes in the ozone layer. Therefore,
using metros is more advantageous, safe and eco-friendly.
The pattern of traffic in a metro is usually very complex because the trains and routes
chosen by a passenger are unknown. Route choice analysis is a study that is related to the
distribution of passengers in the different routes and the trains chosen by them. Dealing with
such abstract and diverse data to infer the required information and modeling of route choice
behavior are two major challenges faced in public transport management.
The emergence of big data analytics has helped to store, process and manage this complex
data, whereas traditional data processing applications are inadequate to handle it. Conduct-
ing route choice analysis is of primary importance to both passengers and metro operators.
For train operators, this analysis will help them to understand how passenger flow takes place
in the metro network and hence improve service reliability. For metro passengers it will be of
great use in trip planning. Indeed, this study can help urban administrators in route sugges-
tions and managing emergency situations.
A metro generally provides its passengers with a smart card facility. A smart card is a
pocket-sized card with an embedded circuit. Every time a smart card is swiped at the station
gate, details of the trip being made with that card are recorded and the monetary value is
stored and debited from the card. This smart card data is used for data collection processes
and hence contributes to the analysis of travel behavior.
In this paper, the probability of passengers choosing a particular route for an Origin to Des-
tination (OD) pairing with multiple routes is shown in Figure 1. Here, big data analytics have
been employed to deal with such vast and complex data. The Hadoop framework has been
preferred for this implementation because it supports batch processing on enormous amounts
of information. The Hadoop framework consists of a distributed file system and a MapReduce
913

Figure 1. Routes in an OD pair.
function. These two elements of Hadoop can be used to enable the storage of huge amounts of
data and for performing parallel processing to save time and improve efficiency.
2 BACKGROUND
Traditional approaches are no longer scalable. The old method of route choice study was to
collect the information from surveys conducted by asking passengers about their routes and
trains. This method was a tedious process and could not yield the best possible results as it
was limited to persons, places and times.
Automated Fare Collection (AFC) systems were then used for the analysis, which gave
broader information regarding the travel pattern of passengers. A drawback related to these
AFC systems was that they did not give any information regarding the train and route chosen
by passengers but just provided the details of the origin and destination stations traveled by
the passengers. In earlier studies, the walking time between the swiping gate and the platform,
and transfer time between platforms, were ignored. These have been considered in this study.
Here we consider three cases and make a comparative study. In one case, route choice
analysis is done using smart card data only; another case study is done using smart card and
timetable data, and the final study uses smart card data, timetable data and MCL data, which
is obtained by means of conductor checks.
3 LITERATURE SURVEY
3.1 Big data

With advancements in technology, enormous amounts of data are generated every second of
every day, as shown in Figure 2. Previously, landline phones were widely used, but advances
in technology have led to them being replaced by smart phones and Android phones, which
have, arguably, made peoples’ lives smarter as well as their phones smarter. There were also
bulky desktops, which were in great demand and were used for processing megabytes of data.
Floppy disks were also in use, and then hard disks evolved and now cloud technology has
emerged for storing huge amounts of data. Enormous amounts of data are generated daily
from the use of smart phones alone. Our every action with a smart phone, such as sending a
video through WhatsApp or any other messenger application, or a post on Facebook chat, is
also generating lots of data. The data generated is not in a format that a relational database
can handle. The volume of data has also increased exponentially and it has become difficult
to deal with such enormous amounts of data. Consider, for example, the case of self-driving
cars that have sensors that constantly record details such as the size of an obstacle and the
distance from it before, finally, considering all these factors and deciding how to react: a huge
amount of data is generated for each kilometer that the car has driven.
914

Figure 2. Advances in technology.
Huge quantities of passive data streams are collected by smart cards, GPS, Bluetooth and
mobile phone systems all over the world. This data happens to be very useful to transport
planners, because of the valuable spatial and temporal information it contains. Big data plays
an important role in storing and processing huge amounts of data in a way that existing
systems cannot. The classification of data as big data is based on volume, veracity, velocity
and value. Volume refers to the exponentially rising amount of data. Variety refers to the
multiple sources from which the data is emerging. The data so obtained are of different kinds
and can be classified as structured data, which includes tables, Excel spreadsheets and so
on, semi-structured data, consisting of cstv, emails, XML files and so on, and unstructured
data, which includes video, images and so forth. Data is being generated at an alarming rate.
Velocity refers to the speed at which the data is produced. Value refers to the mechanisms that
derives correct meanings from the data that is extracted. Veracity considers the difficulty of
extracting such value from the data and helps in handling this.
3.2 Hadoop
Hadoop is one of the frameworks used to handle big data. It consists of a distributed file
system for storing large amounts of data and a MapReduce function to perform parallel
processing on the collected data by assigning work to each processor connected in the net-
work. Hadoop is an open source framework and provides distributed storage and computa-
tion across clusters of computers.
3.3 Smart card data

AFC systems adopt smart card platforms for payment processes in metro systems and trams
all over the world. Only a passenger with a card with sufficient credit can travel on the metro.
The smart card information helps to track the travelers’ behavior in an efficient manner. It
records information regarding the origin station from which the passenger has boarded, the
travel time, the destination station from which the passengers exited, their credit balance, and
the smart card ID. All this information is recorded as and when the passenger swipes the card
in the origin and destination stations. This mechanism also helps to calculate the fare of the
journey and automatically deducts this from the card balance.
There are various benefits to using smart cards for travel purposes. The waiting time spent
at ticket counters can be avoided as the card can be recharged over the internet from any-
where. In addition, it is possible for a passenger to terminate the journey at any time and the
fare will be taken only for the traveled distance. Smart cards are transferrable as they can be
used by friends or family members. All these benefits of smart cards make it more advanta-
geous for travelers to use metros, buses and tram systems.
3.4 Data mining

Data mining is a powerful tool for extracting valuable information from raw data. Data min-
ing, which is also known as knowledge discovery, resembles the gold-mining process in which
915

gold is extracted from the placer deposit; here knowledge is extracted from data. In today’s
scenario, the data is rich but information is poor, so raw data has to be converted into useful
information. The various steps involved in data mining are data pre-processing, data selec-
tion, transformation, data mining, building patterns, data evaluation, and data visualizations.
Some of the principal data-mining techniques are:
• association;
• clustering;
• classification.
Association is a data-mining process that is often used to analyze sales transactions. Items
that are purchased together, for example bread and jam, are associated items and placing
them together in a market yields more profit. Clustering is an unsupervised learning tech-
nique: similar data is grouped into clusters. Classification is a supervised learning technique
performed on the assumption that some knowledge of the data is available. Decision trees,
neural networks, and naive Bayes algorithms are some of the methods involved in this area.
4 METHODOLOGIES
4.1 Analysis using smart card and train timetable data

In this route choice analysis, only the OD pair is taken into account. Several routes exist
between the origin and destination stations. This analysis will help in obtaining the probabil-
ity that a passenger will choose a route in the OD pair as shown in Figure 3.
Two possible journeys can be made by the passengers; either a direct journey from origin
to destination or by making a transfer journey, as shown in Figures 4a and 4b. Consider a
day, I, that is divided into half-hour intervals as {I1, I2, I3, I4, …I48}. The routes chosen in
a particular time interval Ij are considered for the route study.
4.1.1 Datasets
The route choice analysis conducted here utilizes smart card data as well as timetable data.
Timetable data is maintained by the stations to inform passengers of train numbers, train
routes, arrival times, departure times and the metro line name or number. Clubbing both the
dataset the route choice study will be enhanced.
Figure 3. Route graph.
Figure 4a. Direct journey. Figure 4b. Transfer journey.
916

4.1.2 Notations and assumptions
In this system, the smart card data of a passenger, along with the train timetable data, is used
to determine the route information. A probabilistic model has been developed to estimate the
routes and trains chosen by the passenger. Two polynomial distributions are considered here,
that is, the number of trains a passenger waits at his station and at the transfer station. Con-
sider the trip, x, of a passenger; the beginning time of the trip is considered as x.b, the origin
is x.o, the end time is x.e, and the destination is x.d. This information is acquired by joining
together the tap-in and tap-out data from the smart card, as shown in Figure 5. If the trip has
i parts, this means that the trip will need (i-1) transfers. The set of effective routes of an OD
pair is R = {R1, R2, …Rz}. It is assumed that the routes being chosen in a particular time slot
Ij obey the polynomial distribution with parameter αj = {αj1, αj2, …αjz} where ∑α jz = 1.
Given the train timetable and the trip of the passengers X = {x1, x2, …xq} at time slot Ij, it
is possible to calculate αj, which is the maximum likelihood function.
Two assumptions which are usually ignored are considered here: first, the time taken for
the departure of two adjacent trains will be greater than that of the time taken by passengers
to walk between the platform and the charge gates. Second, the majority of passengers leave
the station when they reach their destination.
To calculate αj of an OD pair, all effective routes between the OD pair are needed. Then
the journeys in a particular time slot Ij and the train timetable are used to calculate the maxi-
mum likelihood estimation.
4.1.3 Maximum likelihood estimation

Maximum Likelihood Estimation (MLE) is a technique used for estimating the parameters
of a given distribution, using some observed data. Maximum likelihood estimation is the log
of the likelihood function. The likelihood function is the probability of getting the data when
the probability distribution model for that data is given.
Passengers entering the metro gate at a time slot j in station s and choosing metro line l are
represented by (s, l, j), which is known as the tap-in information, through the charge gate.
Passengers who transfer from metro line l to l’ in transit station s at time slot j are repre-
sented by (s, l, l’, j), which is known as the transfer information.
Figure 5. Route notations.
Figure 6. Processing chart.
917

To calculate the maximum likelihood αj of an OD pair firstly, all effective routes between the
OD pair should be calculated. Secondly, the trips made between the OD pair on a particular
day Ij are considered. By knowing the trips and the routes, all the possible trains and train com-
binations chosen by the passenger can be analyzed by matching the tap-in and transfer infor-
mation. Thus, it is possible to calculate the maximum likelihood of routes chosen by calculating
the probabilities retrieved from θ and β values as shown in Figure 6. Finally, considering both θ
and β values, the maximum estimation value αjz can be calculated using Equation 1:
Maximum likelihood function = Log of likelihood
L(X , Tab, α j ) = log ∑ (α jz ∗ Pr ( xq.e | Tab, xq.b, Rz ) ) (1)

Rzε R
where Pr (xq.e|Tab, xq.b, Rz) represents the possibility that a passenger xq passes through
exit gate at time xq.e on condition of Tab, xq.b and the route chosen, Rz. So, Pr(xq.e|Tab,
xq.b, Rz) can be calculated by summing up the probabilities of all plans.
The smart card data which acts as the input for the route study includes several kinds of
errant data such as missing data, duplicated data and data with logical errors. In the data
pre-processing step this erroneous data will be filtered out and the trips will be extracted
from it. In the next step in generating the route, a dataset from another input, that is, the train
timetable, and certain shortest path algorithms are used. The routes which are not used in the
OD pair are also filtered. Finally, the trips are classified as follows:
• direct route;
• one-transfer route;
• multi-transfer route.
The trips such as the direct route and the one-transfer route help to estimate the value of
θ and β, respectively. With the known θ and β value, the probability of choosing each route
in an OD pair is calculated.
4.2 Analysis using smart card, train timetable and MCL data
The route choice study of passengers provides operators with the opportunity to improve
their passenger service. Analysis using smart card data and train timetable data only gives
information regarding a trip’s origin and destination but not the train information. So, in
order to make an accurate prediction we prefer to consider the smart card data, the train
timetable, and the conductor check data.
The conductor check data is the extra information which is collected from a passenger dur-
ing his or her trip on the metro. Metro conductors check a passenger’s ticket using a mobile
Figure 7. Processing chart.
918

chip card reader. When the passenger’s card is swiped in the reader, information such as the
smart card number, the current time, and the number of the current train will be recorded.
The steps involved in this analysis are route generation, route selection and validation. The
inputs used for this analysis are smart card data, the train timetable, and conductor check
data from earlier days and a specific day d. Routes are generated by using various route-
generation methods such as the Bellman–Ford algorithm or Dijkstra algorithm. In the route-
selection process, the generated route sets are further shortlisted by using certain rules. Then
the validation process is done by checking the selected route sets with the conductor check
data of the specific day d, as shown in Figure 7.
4.2.1 Route-generation methods

Initially, the train timetable data is converted into an event activity network which consists of
the basic network, BN, and the extended network, EN, as shown in Figure 8. The BN consists
of information regarding the trains and the station, and the EN consists of more detailed
information such as the wait time, and transfer stations, along with the train and the station
information. Route-generation methods work by considering factors such as minimum cost,
extended route search, and link learning.
4.2.2 Route-selection methods

After the route-generation mechanism, the multiple routes generated are shortlisted by apply-
ing certain rules to mine the required routes. Some of the selection rules used are:
• First Departure (FD)
• Last Arrival (LA)
• Least Transfers (LT)
• Selected Least Transfers Last Arrival (STA)
Based upon these rules, the routes are filtered and selected from the route set. The FD rule
selects the route whose departure time is closest to the time of check-in. The LA rule selects
the routes that are arriving last and are closest to the time of check-out. The LT rule selects
the route with minimum transfers. STA is a combination between the maximum time dura-
tion route and the LT rule.
4.2.3 Validation
Due to the availability of conductor checks on the specific day d, it is possible to evaluate the
performance of the analysis by comparing the selected route with MCL data and filtering
the required route.
4.2.4 Shortest path algorithms

Routes between an OD pair can be retrieved by referring to the train timetable information of
the stations. The use of shortest path algorithms such as Dijkstra or Bellman–Ford help to find
the minimum-cost routes quickly when applied to a BN or an EN formed by converting the
Figure 8a. Basic network. Figure 8b. Extended network.
919

train timetable. The running time of the Bellman–Ford algorithm is known to be O(|V| × |E|). A
graph with negative weights cannot be executed using the Dijkstra algorithm, so the Bellman–
Ford algorithm is used for this purpose. The Bellman–Ford algorithm is also much simpler
than the Dijkstra algorithm. However, the time complexity of the Bellman–Ford algorithm is
O(VE), which is greater than the Dijkstra algorithm.
4.3 Analysis using smart card data alone

The metro users’ mobility patterns can be studied from different angles using smart card infor-
mation. The passengers’ spatio-temporal travel patterns inside a metro can be inferred from
this. This study will also be helpful in anomaly detection. The same smart card can be used
by a passenger to travel in a metro as well as on a bus or tram. So, two types of smart card
information can be collected from a single passenger’s smart card: metro travel and bus travel.
The additional bus travel information will help in producing clustering with higher precision.
4.3.1 Data-mining tools

K-means clustering is the data-mining technique that is employed for the route study to find
similar groups of data. Clustering is the process of grouping data points into similar clusters.
Data points are assigned to clusters based on the calculation of the square of Euclidean
distance, which is the minimum distance from the data points to the cluster. The data points
are assigned to clusters with low inter-group similarity and high intra-group similarity. This
is preferred because of its simplicity and efficiency.
4.3.2 Temporal analysis

An individual passenger is concentrated upon here and his or her mobility patterns on the
metro and bus are considered. First, we find all the trips that belong to a specific passenger
from the transaction record. This record contains the boarding and exiting information taken
from the smart card used for the bus and metro journey. The second step is pre-processing,
where the passengers who rarely travel are ignored or filtered as they contribute little infor-
mation on the temporal or spatial characteristics. So, only passengers with a card containing
more than six active days are considered. Temporal analysis selects features that have small
values and convey more temporal information. Such features will help improve the scalability
of analysis. The k-means clustering algorithm is used to cluster all passengers based on their
four-dimensional temporal features. As a result, four groups will be formed, such as TGrp1,
TGrp2, TGrp3 and TGrp4, as shown in Figure 9.
The proportion p of active days is categorized as Very Frequent (VF), Relatively frequent (RF),
Frequent (GF), and Least Frequent (LF) on the basis of the study conducted. These proportions
help to classify the TGrp groups into any of these categories based on the rules listed in Table 1.
Based upon the proportions of Table 1, the temporal groups fall into the following catego-
ries as shown in Table 2.
By also incorporating the bus data, the analysis can be more accurately completed. With
bus data and TGrp as centers, all passengers are re-clustered to form four groups BTGrp1,
Figure 9. Temporal analysis chart.
920

BTGrp2, BTGrp3 and BTGrp4. The passengers are then compared and matched with TGrp
and BTGrp. Thus, an inference can be made from this study that some passengers prefer to
travel by metro alone, whereas others prefer to complete their journey by using both metro
and bus, and there are also travelers who use the bus alone to make their journey. A com-
parison of the joint probability distribution of passengers is done and information is derived
about their travel pattern, which is shown in Table 3.
4.3.3 Spatio-temporal analysis

Metro passengers are spatially clustered into four groups using a k-means algorithm on the
four-dimensional spatial features of SGrp1, SGrp2, SGrp3 and SGrp4, as shown in Figure 10.
The conditional probability of SGrp given TGrp is calculated, which shows how SGrp
and TGrp are connected with each other. From this inference we discover that group 1 and
group 2 passengers are regular in their spatio-temporal characteristics. Group 3 passengers
are relatively regular and group 4 passengers are irregular in their temporal and spatial char-
acteristics. There are three group labels for every metro user, which are denoted, respectively,
by TGrp, BTGrp, and SGrp.
Table 1. Proportions.
Group VF RF GF LF
p p ≥ 80% 70% ≤ p < 80% 50% ≤ p < 70% p < 50%
Table 2. Group characteristics.
Group 1 st 2 nd 3 rd 4 th
TGrp1 VF LF LF LF
TGrp2 VF VF LF LF
TGrp3 RF GF LF LF
TGrp4 LF LF LF LF
Table 3. Joint probability distribution (%).
Group BTGrp1 BTGrp2 BTGrp3 BTGrp4 BTGrp5
TGrp1 5.48 7.78 2.54 0.04 15.83

TGrp2 0.00 31.70 0.94 0.00 32.64
TGrp3 0.03 3.55 23.52 0.02 27.11
TGrp4 0.58 5.7 15.17 2.94 24.41
Total 6.08 48.75 42.17 3.00 100.00
Figure 10. Centers of SGrp.
921

Table 4. Class groups.
Class TGrp BTGrp SGrp
Class-1 1 2 1
Class-2 2 2 2
Table 5. Comparative study.
Maximum likelihood
estimation K-means clustering Bellman-Ford algorithm
Estimates the likelihood of Partitions data into clusters and Finds effective paths chosen based
route selection. uses the joint and conditional on minimum path, cost, etc.
probabilities for estimation.
Uses smart card and train Uses smartcard data. Uses smartcard data, train
timetable data. timetable and MCL data.
Applicable to different Applicable to all data but Applicable to network models.
types of data. outliers don’t work well.
Precise estimation. Simple and efficient method. Maximizes performance.
Class-1 represents those who take the metro on one trip and take a bus in another trip;
Class-2 represents those who take the metro in round trips, as shown in Table 4. The compari-
sons of average cost and travel time across five weekdays are shown in Figure. It is clear that
Class-1 is lower than Class-2 in terms of cost and travel time. It was learned that, in Shenzen,
the average cost of taking the metro is higher than taking the bus and, for economic reasons
and if time permits, some passengers will choose the bus.
5 ANALYSIS OF METHODS
A comparison of route choice analysis methods using smart card data is presented in Table 5.
6 CONCLUSIONS
With the development of computer technology there has been a tremendous increase in the
growth of data. Big data analytics is an emerging field in intelligent transportation systems nowa-
days. The objective of this paper was to make a study on route choice behavior using smart card
information, which helps in analyzing the travel patterns of the passenger, their route, and their
train selection. This inference will, indeed, be useful for operators in terms of improving the serv-
ices given to passengers and will also be helpful to passengers in trip planning. A comparative
study has been conducted here to understand the route choice behavior of passengers.
REFERENCES
Agard, B., Morency, C. & Trépanier, M. (2006). Mining public transport user behavior from smart card
data. IFAC Proceedings Volumes, 39(3), 399–404.
Kusakabe, T., Iryo, T. & Asakura, Y. (2010). Estimation method for railway passengers’ train choice
behavior with smart card transaction data. Transportation, 37(5), 731–749.
Trépanier, M., Tranchant, N. and Chapleau, R. (2007). Individual trip destination estimation in a transit
smart card automated fare collection system. Journal of Intelligent Transportation Systems, 11(1), 1–14.
Tsai, C.W., Lai, C.F., Chao, H.C. & Vasilakos, A.V. (2015). Big data analytics: A survey. Journal of Big
Data, 2, 21.
White, T. (2012). Hadoop: The definitive guide (3rd ed.). Sebastopol, CA: O’Reilly Media.
Wu, X., Zhu, X., Wu, G.Q. & Ding, W. (2014). Data mining with big data. IEEE Transactions on Knowl-
edge and Data Engineering, 26(1), 97–107.
922

International Conference on Changing
Cities-Architecture and Energy Management (ICCC)
Architectural applications

Ecological conservation of sites through responsive design:

Case of the Muziris heritage interpretation center
Malavika Gopalakrishnan & S. Surya

School of Architecture, Government Engineering College, Thrissur, Kerala, India
ABSTRACT: This paper outlines the process by which the ecological conservation of the
site of the Muziris Interpretation Center and the maritime museum at Pattanam, North
Paravur was proposed, through site responsive design. The proposed site was unique, with
a major part of the site being an unrealized pisciculture plot (traditional ‘chemmeenkettu’).
The ecological conservation of the ‘chemmeenkettu’ was done after conducting buildability
studies of the site and formulating a proposal for the future realization of the site to its full
potential. The main museum was designed as a floating structure, with minimum damage to
the site and utilizing only a percentage of the waterlogged area. The museum was designed
as a climatically responsive building by using renewable energy systems in floating structures.
1 INTRODUCTION
The Muziris Heritage Project aims at reinstating the historical and cultural significance of
the legendary port of Muziris. The region is dotted with numerous monuments of a bygone
era that conjure up a vast and vivid past, and the Muziris Heritage Project is one of the big-
gest conservation projects in India, aiming to preserve a rich culture that is around 3,000
years old. The material evidence unearthed at the excavation site, located about 25 km north
of Kochi, points to the possibility that Pattanam may have been an integral part of the legen-
dary Port of Muziris, and thus the interpretation center at Pattanam helps to promote aware-
ness and understanding of the cultural distinctiveness and diversity of Muziris.
The Muziris, located along the west coast of Kerala, has the most productive waters in the
world. It paved the way for the supreme diversity and abundance of both fishes and fisher
folk. The indigenous and conventional methods of fishing are rapidly declining and are in
need of conservation. Brackish water fish farming and cage fish farming are unique methods
of culturing fishes and these eco-friendly methods of fishing provide livelihood security for
the fisher folk in the region, as well as enhancing its biodiversity by supporting varied eco-
systems. The proposed site for the interpretation center is one such, traditionally known as a
‘chemmeenkettu’.
2 ARCHITECTURAL CONCERNS OF THE PROJECT
2.1 Understanding the tangible and intangible elements of Muziris heritage

The Muziris Heritage Site is an outstanding example of the buildings, archaeological sites
and landscapes that illustrate a significant stage in the history of Kerala. The area bears an
exceptional testimony to a culture that is fast disappearing. The built heritage of Muziris is
extensive and spreads across the site. While the physical remains of Muziris are outstanding
and constitute a unique ensemble, there are also intangible associations and traditions that
form an important part of Muziris. The practice of fishing is thousands of years old. One
such practice is the unique Chinese nets and brackish water shrimp farming.
925

2.2 Recreating their lost identity
The Kottapuram and Paravur regions were thriving commercial centers, whose fabric had a
variety of Portuguese, Dutch and traditional Kerala influences. The market streets portray the
pressures and problems that it faced in earlier times. The culture and the traditions are the sig-
nificant factors that make it a unique destination. The development that took place, in terms
of scale, design and typology, was unsympathetic to the original character of the region, affect-
ing the identity of the place. Reviving the lost traditions, and at the same time creating future
opportunities, are the key aspects of recreating the glory of this once prestigious port city.
3 LOCATION AND CHARACTERISTICS OF THE SITE
The site is located at Pattanam, North Paravur, in Kerala, with an area of 49 acres and 92.84
cents. It is located about 1 km from the Kodungallur–Paravur route. The site is along the
banks of the Kollam–Kottapuram waterway (National waterway III), encompassing the west
and south side of the site.
3.1 Features of the proposed site

The majority of the site (68%) is an unrealized brackish water shrimp farming plot in a total area
of 50 acres. The site is also surrounded by Chinese fishing nets and local fishing centers. The site is
quiet and serene, away from the city. Even though the location is remote, it can be easily accessed
by the upcoming Muziris heritage circuit. The site is also near to the Pattanam excavation site.
In Kerala, there is a long history of the traditional system of salt water aquaculture being
practiced in the seasonal Pokkali fields and the perennial fields lying adjacent to the coastal
and backwater areas. In these fields, seed shrimps and fishes are allowed to enter through
tidal water and are then trapped. After a short duration of growth, they were periodically
harvested during full and new moon periods, from November–April.
The site has an average depth of 4–5 ft. (1.2 m–1.5 m) and a maximum depth of 2 m. It is
also known locally as ‘chemmeenkettu’. The fishing season starts from November and lasts
until April, and the harvest is done at ‘thakkam’ (7 pm-4 am). The water in the chemmeen-
kettu is recycled from the river during the day.
3.2 Architectural character of the site

The architectural pattern around the site is a traditional Kerala style with inspiring forms of
traditional architecture and pristine backwaters. There is also a significant colonial influence,
with elements of Dutch and Portuguese architecture. The material typology has sunburned
and laterite bricks with non-imposing sloping roofs done in Mangalore tiles and sheet roof-
Figure 1. North Paravur site map.
926

Figure 2. Google earth image showing site.
Figure 3. Site image. Figure 4. Site image.
ing. Some private residences around the site are mostly MIG (Middle Income group) and
LIG (Low income Group) occupancy with single storied flat roofs.
3.3 Neighborhood context

Pattanam means ‘town’ in the local language of Malayalam and it is located 8 km south of
Kodungallur and 25 km north of Kochi in the Ernakulum District in the southern Indian state
of Kerala. The site is located about 1 km from the Pattanam excavation site. Paravur market
and synagogue are also in close proximity. The Sahodaran Ayyapan museum is located opposite
the site. Recent archaeological excavations have unearthed signs of early Roman trade, which
was part of Muziris that flourished during the reign of the first Chera dynasty of South India.
National Waterway No 3 is a 168 km stretch of this inland navigational route located in
Kerala, India and runs from Kollam to Kottapuram. It was declared a National Waterway
in 1993, and connects the industrial centers of Kochi to the Kochi port. The majority of
those inhabiting the region were engaged in traditional industries, such as coir, cashew, brick-
making and fishing.
4 ECOLOGICAL CONSERVATION OF THE SITE AND RESPONSIVE DESIGN
4.1 Buildability studies of the site

The unique nature of the site required the conducting of buildability studies in order to
determine a suitable location for the project with minimum intervention. The key aspect of
927

Figure 5. Pattanam and nearby landmarks.
Figure 6. Site showing land area and waterlogged area.
Figure 7. Site zoning.
928

conservation is centered on not undertaking any landfilling or affecting the shrimp farming
practice. The entire site is demarcated via a bund system for pisciculture, but due to a lack
of resources and poor maintenance, only 50% of the site is utilized. There is no commercial
fish farming and, during the off season, the site remains as a waterlogged area, not realizing
its full potential.
To understand whether alternative construction techniques, such as stilt systems, pier and
floating systems, were possible, a detailed ‘depth analysis’ was undertaken for each bund.
Figure 7 shows the various depths in each part of the site. As a floating system is the most
Figure 8. Depth analysis of the site.
Figure 9. Site analysis (a) views to the site (b) views from the site (c) access and entry to the site (d)
circulation and road hierarchy within the site.
929

sustainable method to use for minimum disruption to the natural setting, the minimum depth
required was 1 m. As seen from the studies, the area toward the left had an average depth of
3 m and, thus, a floating system was proposed for the museum.
Further, the views, access, entry, circulation and road hierarchy were analyzed to under-
stand the suitable orientation and placement of the buildings.
4.2 Architectural language achieved: Site responsive design considering local climatic factors
Considering the tropical climate, natural ventilation, a sloping roof, long overhangs, passive
and active cooling through orientation, and vertical louvers were achieved.
The natural flow of air is achieved by having full length vertical louvers on the river-facing
side. The building is oriented so that the longest side is facing the wind direction.
The building is L-shaped with courtyards in the middle, which will act as a pressure point
for catching the wind. Further, clerestory windows are also provided to allow the hot air to
rise and flow outside, thus maintaining a circular air circulation system within the building.
5 ARCHITECTURAL STRATEGIES TO ADDRESS THE CONCERNS
5.1 Relevance of floating architecture in sustainable design

Floating architecture can be defined as a building for living/working that floats on water with a
flotation system and is moored in a permanent location. Floating architecture on water has been
emerging as a sustainable alternative around waterside regions, and floating architecture can be
regarded as one of the most sustainable building types if proper sustainable factors are applied.
Floating architecture could be a future solution for the current problems in many districts, cities
and landscapes. Such problems can especially be seen in the need for additional housing areas
and construction grounds in some countries in Europe and Asia, as a result of the growing
population and/or the slowly rising sea levels in the context of worldwide climate change.
5.2 Construction and design of floating structures

Steel pontoon float technology is used to keep the museum in its proper position while float-
ing on the river. While the buildings must be designed with enough structure for lateral and
gravitational loads, they must also be stabilized in the moving river. The museum floats on
a pontoon system, similar to the floating runways used for airplanes, and is held to the river
bed by anchoring with the help of a pile system.
Additionally, the angle and shape of the chains are controlled by Dynamic Positioning Sys-
tem technology to ensure stability. Because the buildings are floating, the construction process
requires the prefabricating of the structure of the museum on the riverbanks, and then launch-
ing them in the river using a field of rollers as casters, similar to launching a boat into the water.
Construction of the structural frame is in steel to enable large spans. They are then covered
in mesh wrap and given cement plaster and coated with corrosive resistant paint. For the
walls, lightweight materials called V-boards are used.
Figure 10. Types of steel pontoons.
930

Figure 11. Steel pontoons assembled together.
Figure 12. Site layout.
5.3 The floating mechanism

Airbags are placed, each measuring 2 meters in diameter and 12 meters in length, under the
giant buoy. The airbags are made of a special rubber that can support a 2,000 ton buoy, and
will enable it to keep afloat.
The buoy is then embedded to the river bed using anchors (coconut piles) with metal chains
to prevent it from moving or washing away. The chains are designed so as to automatically
tighten when the water level increases.
The key principle is to balance the weight of the building to that of the water displaced. If
the weight of the building is less than that of the water displaced, it will float.
5.4 Use of renewable energy in floating structures

5.4.1 Solar energy
Solar photovoltaic power generation has long been seen as a clean energy technology that
draws upon the planet’s most plentiful and widely distributed renewable energy source—the
sun. Cells require protection from the environment and are usually packaged tightly in solar
panels. The main museum receives its energy from the solar farm, proposed at the site, mak-
ing it more sustainable.
5.4.2 Hydrothermal cooling

Hydrothermal temperature control involves pumping the river water from the Kollam–
Kottapuram waterway through a vein-like system in the wall and floor, and this cools the
building structure down by 15 degrees Celsius (from 45 degrees to 30 degrees), reducing the
cost of cooling by around 40 percent. Humidity control is usually considered to be the most
expensive aspect of environmental control. The optimum temperature and humidity for a
museum is 18–20 degrees Celsius and 45–50% respectively.
931

Temperature and humidity control has been achieved to a great extent with the help of
climatically responsive design. In addition to that, dehumidifiers can be placed in every
gallery.
6 CONCLUSION
The Muziris Interpretation Center and maritime museum is a conscious effort to achieve
a site responsive design, stressing the importance of sustainable and eco-friendly architec-
ture. The proposal is aimed at inducing a deep appreciation for the lost heritage of Muziris,
through understanding the context of the site and celebrating the essence of its uniqueness.
The aspect of conservation leads the project by challenging the conventional construction
practices.
One of the most challenging aspects of the design was the distinctiveness of the site. As the
site is a pisciculture plot, the focus was on nurturing and protecting the natural setting. The
site buildability studies and analysis revealed that a larger part of the site was better left as it
is, and this can be developed further in the future for aqua tourism. On the remaining plot,
the museum was proposed as a floating structure with minimum intervention to the site. Each
gallery was visualized as an island floating amidst the tranquil setting, connected through
bridges. This not only makes the museum stand out, but elevates the visitor experience.
An ecological and climatically responsive design practice ensures maximum sustainability
and retains the natural setting of the site. Further, renewable energy sources, such as solar
and hydrothermal energy, are also incorporated in the design, in order to uplift the design.
REFERENCES
Adler, A.D. (1998). Metric handbook planning and design data (3rd ed.), Routledge; 5 edition (5 March
2015).
Cherian, P.J., Selvakumar, V., and Shajan K.P., 2007a “The Muziris Heritage Project: Excavations at
Pattanam- 2007”, in Journal of Indian Ocean Archaeology, New Delhi, Vol 4, pp. 1–10.
Cherian, P.J. (2014). Unearthing Pattanam, catalogue for 2014 exhibition. National Museum, New
Delhi.
Habibi S (2015) Floating Building Opportunities for Future Sustainable Development and Energy Effi-
ciency Gains. J Archit Eng Tech 4:142. doi:10.4172/2168-9717.1000142.
Kuriakose, B. (2009). Conservation development plan for Muziris heritage sites (Consulation draft),
Chennai.
932

Design strategies for daylighting in tropical high rises
Nishan Nabeel, Anju John & A.K. Raseena

DG College of Architecture, Chelambra, Kerala, India
ABSTRACT: Daylighting is one of the measures adopted to take advantage of the climate
and environment in designing buildings. Here emphasis is given to the strategies that can be
adopted to bring daylight effectively to the tropical high rises thereby reducing the energy
consumption due to artificial lighting. The tropical climate differs substantially from condi-
tions of other climates. These differences should be taken into account while designing the
facades of high rises in this region. This paper mentions some of the terms regarding the
daylighting components, daylighting systems, principles and their analysis in the context of
tropical high rise. It also analyses some of the case studies and derives design strategies that
can be adopted for daylighting in tropical high rises.
1 INTRODUCTION
Daylighting is the controlled admission of natural light, direct sunlight and diffused sky-
light into a building to reduce electric lighting which in turn saves energy of the building.
If used efficiently, it helps to create productive environment for the building occupants as
well as reduce one third of the total building costs. Daylighting design focuses on how to
provide enough daylight to an occupied space without undesirable side effects like heat gain
or glare. Therefore it balances heat gain or heat loss, variation in daylight availability and
glare control.
The main aims in daylighting a building are to (1) get significant quantities of daylight as
deep into the building as possible, (2) to maintain a uniform distribution of daylight from
one area to another, and (3) to avoid visual discomfort and glare1.
2 DAYLIGHT MEASUREMENT
2.1 Daylighting data

The quantity and quality of the light generated by all the sources at a particular point is dif-
ficult to calculate. Luminance is the first factor to be considered. The intensity of illumination
from direct sunlight on a clear day varies with the thickness of the air mass it passes through—
a function of the angle of the sun with respect to the surface of the earth2. Light is intense at
noon than at sunrise and sunset, and intense at lower latitudes than at higher ones.
It is difficult to codify the variations in the sky luminance caused by weather, season and
time of the day. Several standard sky models have been produced to meet this difficulty like
Uniform Luminance Sky Distribution, CIE (International Commission for Illumination)
Standard Overcast Sky Distribution and Clear Blue Sky Distribution. For apertures fac-
ing away from the sun, reflected light is an important source of indoor daylight. External
reflected light includes the light reflected from surrounding surfaces like ground, water, veg-
etation, other buildings.
1. Lee Jin You, Roger, Lee Ji Hao, Theophilus, Sun and architecture, heavenly mathematics.
2. The European Commission Directorate-General for Energy, Daylighting for buildings.
933

2.2 Daylight factor
(DF) is the ratio of the light level inside a structure to the light level outside the structure. It
is defined as:
DF = (Ei/Eo) × 100% where, Ei = illuminance due to daylight at a point on the indoors
working plane, Eo = simultaneous outdoor illuminance on a horizontal plane from an unob-
structed hemisphere of overcast sky. To calculate Ei, requires knowing the amount of outside
light received inside of a building. Light can reach a room via through a glazed window, roof
light, or other aperture via three paths:
• Direct light from a patch of sky visible at the point considered, known as the sky compo-
nent (SC),
• Light reflected from an exterior surface and then reaching the point considered, known as
the externally reflected component (ERC),
• Light entering through the window but reaching the point only after reflection from an
internal surface, known as the internally reflected component (IRC). The sum of the three
components gives the illuminance level (typically measured in lux) at the point considered:
Illuminance = SC + ERC + IRC3
2.3 Sky component

Sky component depends directly on the position of the sun in the sky and the angle in
which the sun’s rays enter the building. The sun’s path varies with location (local lati-
tude), rising and setting position (based on the time of the year), duration of the day
and night.
To measure the angle of the sun in its motion across the sky, we need to take its alti-
tude and azimuth reading. Altitude is the angular distance above the horizon measured
perpendicularly the horizon. It has a maximum value of 900 at the zenith, which is the
point overhead. Azimuth the angular distance measured along the horizon in a clockwise
direction4.
Figure 1. Azimuth and altitude. Figure 2. Angle of the sun in summer and
Source: Lee Jin You, Roger, Lee Ji Hao, winter.
Theophilus, Sun and architecture, heavenly Source: http://www.build.com.au/window
mathematics. orientation-and-placement.
3. https://en.wikipedia.org/wiki/Daylight_factor.
4. Lee Jin You, Roger,Lee Ji Hao, Theophilus, Sun and architecture, heavenly mathematics.
934

Figure 3. Singapore sun path.
Source: H. Schepers, Mc Clintock, J. Perry, Daylighting design for tropical facades.
3 POSITION OF SUN IN THE TROPICS—SINGAPORE AS AN EXAMPLE
Singapore located on the equator has on average 6 hours of sunshine in a day and high pro-
portion of clear and partly clear blue skies. The sun path in the figure below shows that the
sun is typically high overhead and traces a path that is almost directly East to West5. Hence
to block the direct solar penetration on North and South facades, horizontal shading may
be effective. The East and West facades will be exposed to low angle sun for few hours in the
morning and evening which need not be treated.
A clear sky indicates high luminous intensity in the zone directly adjacent to the sun.
Hence skylight from a clear sky often penetrate deeper into a building which may also cause
glare to the occupants that have direct view to the sky. So in such regions, high light transmit-
ting glazing should be used in combination with the blinds.
There are many new design opportunities which respond to climate and solar path of
the tropics. A good understanding of sun and sky luminance is necessary to evaluate these
opportunities. The statistical information on the sky distribution of tropics is very little and
more research is needed to be carried out in this area.
4 DESIGN STRATEGIES USING DAY LIGHT
Design strategies for daylighting includes daylight optimized footprint, efficient window
design, high performance glazing, passive or active skylights, tubular daylight devices, day-
light redirection devices, solar shading devices, daylight responsive electric lighting controls,
daylight optimized interior design which includes furniture design, space planning and room
surface finishes.
4.1 Building footprint

• The building should be oriented along the East West direction so that the longer façade is
well exposed to the north light.
• Deviation from the south should not exceed 15 degree in either direction for best solar
access and ease of control.
• The floor depth should not be more than 60 ft (18.2 m)
5. H. Schepers, Mc Clintock, J. Perry, Daylighting design for tropical facades.
935

Figure 4. Building footprint.
Table 1. Minimum glazed areas for view when windows are restricted to one wall.
Depth of room from outside wall (max) Percentage of window wall as seen from inside
< 8 m 20%

8–11 m 25
11–14 m 30
> 14 m 35
4.2 Efficient window design

One of the principal challenges of daylight design is achieving good daylight levels away from
the external walls as daylight factors decrease as distance from windows increases. The daylight
penetrates significantly about 4 to 6 m from the external walls in a conventional building. Gen-
erally a room will be adequately lit to a depth 2 to 2.5 times the height of the window from the
floor, so taller rooms can be daylit to greater depth. Simply increasing overall window size will
not be productive if it raises light levels close to the window more than it raises light levels deeper
in the room. So the reflectance values of internal materials should be as high as possible.
To design windows intended to view outdoors, the factors like nature of exterior land-
scape, size and proportion of the interior space, and the positions and mobility of the people
who occupy it should be taken into consideration. Minimum areas of glazing for rooms
which are lit from one side only as shown in Table 1. This total area should be distributed so
as to provide some view from all occupied parts of the room6.
Daylight optimized fenestration where the window serves the both functions of daylight
delivery and view to the occupants is another option. This type of window has two compo-
nents—daylight window and view window. Daylight window starts at a height of 2.2 m above
the finished floor with 50% to 70% transmittance and view window below that with a 40%
transmittance. Efficient window design will be different with increasing levels as the angle of
the sun differs at different heights.
4.3 High performance glazing

High performance glazing admit more light and less heat when compared to the typical window
which allows daylighting without negative impact on building cooling load in summer. It is achieved
through spectrally selective films and usually consists of a double pane insulated glazing unit.
4.4 Passive and active skylights

Passive skylight will be covered with clear diffusing medium which allows light to penetrate.
It can be double layered to increase the insulation. Active skylights consist of a mirror system
that tracks the sun and is designed to increase the performance by channeling sunlight down
into skylight well.
6. The European Commission Directorate-General for Energy, Daylighting for buildings.
936

Figure 6. Atria in high rises.
Source: The European Commission Directorate-
General for Energy, Daylighting for buildings.
Figure 5. Daylighting devices.

Source: The European Commission
Directorate-General for Energy, Day-
lighting for buildings.
4.5 Other daylighting devices

The Fig 5 shows some of the daylighting devices used in typical buildings. Well, in high rises
only a few of the devices can be used. For example, top lighting is mostly preferred for low
rises but they can be provided on top of atrias in high rises.
But a continuous atria might give a tunneling effect for the occupants. Hence instead of
continuous atrias, splitted atrias can be provide which enables both side lighting and top
lighting as shown in Fig 6.
All the types of sidelighting can be provided on the facades of high rises keeping in mind
the building orientation. Each façade may be treated differently according to the intensity of
daylight and glare factor.
4.6 Shading devices

Shading devices are very much required in high rises to shade the large amount of façade
exposed to the sun. They play a major role in creating relaxed inner surroundings that is cool
in summer and warm in winter. Discount of 5% to 15% in annual cooling electricity intake
has been reported when shading devices are used. They also help to differentiate one façade
from another. It helps in maintaining comfortable indoor environment, reduces heat gain and
improves the daylighting quality of building interiors.
937

Figure 7. Shading devices.
Source: Canada-Daylighting-Guide-shading-5.jpg.
Trees Planted serve as shading devices and beautifies landscape and provides oxygen to
the occupants. Internal shading devices also help to create a sense of privacy. One of the dis-
advantage of using shading devices is that it obstructs outdoor views for occupants in some
cases. The solar geometry explains that the publicity of each facade to the sun is specific, and
varies through orientation.7 Each façade should be treated differently. For example, facades
facing north in the northern hemisphere would not need shading devices as solar penetration
is restricted to only few months of summer. Whereas in south elevation, solar penetration
should be controlled. Horizontal shading devices above windows are best suited here. The
length of the projection depends on the height of the window and the attitude of elevation of
the solar at sun noon. It should be designed in such a way that it absolutely get rid of solar
penetration in summer time and allows complete solar penetration in winter.
Steps to consider while designing shading device involves 1 Understand the sun path of the
environment 2 Select the shading type – Horizontal, -vertical, -egg rate 3 Identify category-
Fixed shading devices, – Adjustable shading device, – Movable shading, device-Dynamic
shading, device-Automatic shading device 4 Calculate the design dimensions – To under-
stand horizontal and vertical shadow angles.8
4.7 Daylight responsive lighting controls

It consists of continuous dimming controls or stepped controls by sensing the availability of
natural daylight and the occupant. The photocells in the light fixtures sense he available light
and dim or turn off the electric light in response which enhance the energy reduction.
7. Mustapha Adamu Kaita, Dr. Halil Alibaba, research paper on Shading Devices in High Rise Build-
ings in the Tropics.
8. Mustapha Adamu Kaita, Dr. Halil Alibaba, research paper on Shading Devices in High Rise Build-
ings in the Tropics.
938

Table 2. Shading strategy by window location.
Window orientation Shading strategy
North Usually not needed

South Overhang, horizontal louvers, trellis over window
East/West Vertical louvers, horizontal slats, deciduous trees
Source: Canada-Daylighting-Guide-shading-5.jpg.
Figure 8. Seasonal performance of shading, redirection devices.

Source: Whole building design guide by Institute of building science, California.
4.8 Daylight optimized interior design

Furniture layout in an interior space can regulate the incoming of daylight. For example in
case of offices, arrangement of the cubicle partitions play important role in achieving contin-
uous daylight in all seasons. Heights of the partitions running parallel to south façade will be
limited and positioned at a distance from the façade, which enhances the use of smaller shad-
ing devices. A minimum of enclosed offices are provided to allow penetration of daylight in
combination with highly reflective surfaces to bounce and distribute the redirected daylight.
5 EVALUATION OF CASE STUDIES
After the literature study, how these strategies have been adopted practically has to be evalu-
ated. For the purpose, 5 different tropical high rises have been selected and analysed in terms
of daylighting strategies adopted.
5.1 The Hansar, Bangkok, Thailand (Residential)

Building footprint: Sky gardens are introduced at every 5th floor across east and west eleva-
tions allows ample penetration of low density light. Daylighting for all apartments has been
ensured by organising the units around a central core and courtyards coupled with vertical
slots throughout the building heights.
Shading devices: Balconies acts as shading devices in most apartments across their front-
age. This works in combination with delicate sunshades to keep the interior cool at all times.
A double layer system was devised to provide shading. As high rise buildings in the tropics
gain almost their entire solar load from vertical surfaces rather than its roof, shading of walls
is much effective than of roofs.
Façade treatment: A porous façade in the form of a metal mesh screen has been designed
to hang off projected ledges at every floor serving as the buildings outer skin. This façade
helps in regulating light and air throughout its fully vertical surfaces.
Window design: These external sunscreens overlap with the inner window layer and also func-
tion as privacy screens creating a foreground that frames the city views from within the units.
939

Figure 9. Typical plan of the Hansar.
Source: International journal for Tall build- Figure 10. External sunscreens at Hansar.
ings and urban habitat. Source: International journal for Tall build-
ings and urban habitat.
Its staggered placement creates visual interest from exterior while neatly concealing a/c
condenser units and services in the background. The sun screens are specially coated with
metallic bronze colour.
5.2 Marina bay sands, Singapore, Malaysia (Hotel)

There are three concrete towers. Each concrete tower hotel is designed at a height of 55 sto-
ries. Spanning across the top of the three towers is a 3 acre sky park—a space framing large
urban windows between the towers.
Building footprint: The structure is oriented in North South direction with longer facades
facing East and West. Here importance was given to the views to the sea.
The three void spaces are connected by one continuous and conditioned glazed atrium which
allows large penetration of light to the restaurants, retail spaces and a public thoroughfare.
High performance glazing: Largest amount of heat gain occurs in the west façade. The
design solution implemented was custom double glazed unitized curtain wall.
Shading devices: Perpendicular to the façade, glass fins were installed to provide shading.
The east façade utilises deep planted terraces which follows the sloping radial geometry of
the buildings profile. The planter’s help to create microclimate cooling and the deep over-
hangs of the balconies shade the hotel rooms from direct sun.
5.3 Kohinoor square, Mumbai, India (Mixed use)

Daylighting devices: The main building houses spacious lobby and double height landscaped
Sky Gardens and a double height terraces with floor to ceiling glazing on every alternate
floor to act as tranquil and refreshing breakout zones that allows entry of daylight to the
interiors. This follows the concept of splitted atrias discussed earlier.
High performance glazing: The façade consists of faceted unitized aluminum curtain walls
with provisions for high performance double glass façades on the tower.
5.4 Kanchanjunga apartments, Mumbai, India

Building footpint: Kanchanjunga apartments are oriented in East West direction to face the sea.
Window design: Maximum apertures are provided in these facades. In the South and North
facades, apertures in the form of balconies are only provided, to allow adequate amount of
daylight along with shading from the hot sun
940

Figure 11. View of Marina sands bay.
Source: International journal for Tall Figure 12. East façade.
buildings and urban habitat. Source: International journal for Tall
buildings and urban habitat.
Figure 13. Double height terraces.

Source: Kohinoor square, a multifaceted development—research paper by Sandeep shikre.
Figure 15. South, West, North and East Elevations.

Source: https://identityhousing.wordpress.com.
Figure 14. Daylight distribution.

Source: https://identityhousing.wordpress.com.
Figure 16. Air wells between towers. Figure 17. South façade detail.

Source: 2013 On Site Review Report on Source: 2013 On Site Review Report on Met tower
Met tower by Zainab Faruqui Ali. by Zainab Faruqui Ali.
941

East and West facades are enhanced with lot of apertures both in the form of windows and
balconies to enable the entry of low intensity morning and evening daylight.
5.5 Met tower, Bangkok (Residential)

Building footprint: The staggered configuration of the plans coupled with the separation of
three towers by air wells provides ample natural ventilation and daylight indoors.
Daylight devices: Balconies allows daylight penetration along with shading in the south
façade. The apartments in the Met are naturally ventilated, with access to light and air on all
four sides because the tower effectively has no circulation core, in the traditional sense.
Shading devices: Expanded metal mesh was selected for the sun-shading canopies of the units.
6 CONCLUSION
From the above study, we can understand that, in the case of high rises, design strategies for
day lighting has to be adopted mainly in the building footprint and the façade treatment.
Although we can find a lot of day lighting devices, the ones that we can use in high rises are
limited. For example, courtyards with skylights are one of the major day lighting devices
used in low rises which is not possible in high rises due to its tunneling effect.
The day lighting devices that can be adopted in high rises of tropical region mainly include
the use of balconies, sky gardens, air wells, garden pockets splitted atrias etc... These strate-
gies have many functions like daylight penetration, sun shading, maximize outdoor views,
acts as relaxation space and provides natural ventilation too. More than efficiency of window
design, balcony design is emphasized in high rises.
Another point to be considered while designing the shading devices are that it should be
designed keeping in mind the angle of the sun that changes with increasing height in high
rises. The shading devices adopted in 4th floor will be different from that on the 60th floor.
From the case studies we can find that another important strategy used in high rises are
the high performance glazing mainly in south and west facades to allow daylight penetration
without solar gain. North south orientation which is desirable for daylight penetration is
given less importance and the buildings are oriented in such a way that outdoor views like
view to the sea, rain penetration etc … are given importance like in Marina bay sands and
Kachenjunga apartments.
We can also find splitted atrias as in case of Kohinoor apartments which can be used effi-
ciently as a common entertainment space as well as a day lighting device.
REFERENCES
Adamu Kaita, Mustapha,. Alibaba Halil, (2016, October) Shading Devices in High Rise Buildings in the
Tropics, International Journal of Recent Research in Civil and Mechanical Engineering, Vol. 3, Issue
2, pp: (37–46).
Ander, Gregg, (2016) Daylighting, Whole building design guide, National Institute of Building sciences.
Daylight_factor (2013, July) retrieved from https://en.wikipedia.org/wiki/Daylight_factor.
H. Schepers, Mc. Clintock, J. Perry; Daylighting design for tropical facades.
Lee Jin You, Roger; Lee Ji Hao, Theophilus; Sun and architecture, heavenly mathematics.
Retrievedfrom https://identityhousing.wordpress.com.
Wong, Mun Sum (2012), Case study—The Hansar, Bangkok, CTBUH journal.
Shikre, Sandeep,Kohinoor square, a multifaceted development.
Sun orientation-and-placement (2013) retrieved from http://www.build.com.au/window Sun
orientation-and-placement.
The European Commission Directorate-General for Energy, Daylighting for buildings.
942

Internalizing the concept of sustainability—redefining the

curriculum of building construction in architecture schools
Om Prakash Bawane
R V College of Architecture, Bangalore, India
ABSTRACT: Environmental consequences of contemporary construction practices have

been scientifically established. Construction activities constitute about 40% of global CO2
emissions. The seriousness of the issue has given rise to the promotion of sustainable con-
struction technologies worldwide. It is imperative that curriculum content in the subject
‘Construction Technologies and Materials’ is critically examined to ensure that the issues
of sustainability are adequately addressed. The coursework in building construction must
help internalize the concept of sustainability in the architecture profession. Sensitizing the
future generation of architects toward environmentally friendly construction practices would
go a long way in mitigating the adverse impacts of construction on the environment. This
paper is an attempt to examine the course content in the subject ‘Building Construction and
Materials’ typically adopted at undergraduate programs in architecture schools in India, in
the context of sustainability.
Keywords: Sustainability, Construction, Building Materials, Environmental Impact, Course

Contents
1 INTRODUCTION
The subject of ‘Building Construction and Materials’ occupies an important place in the
under graduate architecture curriculum. In terms of the number of teaching hours and cred-
its, the subject is placed second in order next to the core subject of architectural design.
A sequential exposure to the process of construction through seven to eight semesters of
structured syllabi helps students appreciate various aspects of the constructability of a design
idea. Construction is the process and means of transforming the architectural ideas into a
built product. In architectural projects, methods and materials of construction could be vital
elements in shaping the overall design concept. Architects are trained to be conscious of this
fact and exercise their prerogative on the selection of construction systems and materials
that would enhance the architectural quality of the build environment. Ironically, any act
of construction has a bearing on the natural environment and ecology. The design idea and
visualization of the overall build environment are what finally influence the architects’ deci-
sions on methods and materials of construction.
1.1 Weightage to the subject

Architecture curriculum at undergraduate level in India is largely guided by the ‘Minimum
Standards of Architectural Education’ laid down by the Council of Architecture [1]. In the
recommended allocation of teaching hours, ‘Building Construction and Materials’ has been
assigned about 504 hours, which makes a significant share of approximately 12% of the 4,176
total teaching hours.
A closer look at the architecture curriculum of ‘Building Construction and Materials’ in
architecture schools across India would establish the importance of this subject in the training
943

of prospective architects. The curriculum introduces the subject right from the first semester
with few exceptions in some schools where the subject is taught from the second semester
onwards. The contents of the course are typically spanned over five to seven semesters. In
the scheme of teaching, the subject on average is assigned six hours per week. The scheme of
teaching of all sample institutions indicates that in terms of weekly teaching hours, ‘Building
Construction and Materials’ is placed second after the core subject of architectural design.
In the absolute marking system, the subject enjoys second highest weightage, which is very
explicit.
The weightage assigned to the subject in terms of the number of teaching hours and marks
highlights its overall importance in the architecture curriculum. There exists a close con-
nectivity between the proceedings of design studios across the semesters and the sequence
in which the various concepts of the construction systems are introduced to the students. It
is imperative for prospective professionals to appreciate the implications of the construction
system on architecture and also the intricacies of integrating the construction and design
ideas.
2 UNDERSTANDING SUSTAINABILITY
The word sustainability finds it origin in the Latin word sustenere meaning to hold, bear,
endure, etc. [5]. The term sustainable has a much philosophical connotation meaning any
concept, product or process that would not undermine the capacity of the earthE’s ecosystem
to maintain its essential functions. In context of the definition of sustainable development
advocated by the World Commission on Environment and Development, sustainability is
perceived as a holistic concept encompassing social, environmental and economic aspects
also referred to as the three pillars of sustainability. The training of architects needs to recog-
nize the most sensitive aspect of sustainability, namely the environment, since construction is
one of the major consumers of natural resources and at the same time a major pollutant of
the natural environment.
2.1 Construction and environment

The construction industry accounts for 40–50% of global energy usage, nearly 50% of world
water usage and around 60% of the total usage of raw materials activities (Wilmot Dixon,
2010). Consequently, at a global level construction activities constitute 32% of global CO2
emissions, 40% of drinking water pollution, 30–40% of greenhouse gases and 40–50% of
solid waste generation.
2.2 Sustainable construction

Sustainability is still a new concept in construction and yet to be accepted as an integral part
of decision making processes. Considering the enormity of adverse impacts of construction
on the environment, it is imperative that construction in its totality embraces sustainability.
There are little options except sustainable construction practices. Sustainable construction
is a holistic approach that considers the reproduction capacity of nature with regard to the
materials drawn from the natural stock, energy spent on making a finished product available
on the construction sites, impacts of the construction process on the environment and finally
the life cycle cost.
3 COURSEWORK IN CONSTRUCTION
The contents and structure of coursework in the subject of ‘Building Construction and
Materials’ are found to be long-established in most institutions. In certain universities
the sequence of exposure aligns with the sequence of onsite construction processes, for
944

example, they begin with study of the different types of foundations (substructure), walls
and frames (superstructure), including other elements, such as stairs, floors and finally
the roofing systems. The other approach is material-based formulation, which begins
with construction with basic materials, namely clay, stone, timber at lower semesters, fol-
lowed by construction in concrete, steel and aluminum. The approach toward delivery of
contents appears rather mechanically sequenced and devoid of any stance on issues of
sustainability.
3.1 Orienting the course toward sustainability

The natural resources are finite and bound to exhaust if their rate of consumption exceeds
the reproduction capacity of the earth. Sources of building materials extracted form earth’s
strata are depleting at a rapid pace and consumption of green stock like timber has exceeded
the reproduction capacity of our forests. It is therefore crucial to sensitize the students about
sustainable consumption of natural resources. Energy-related air pollution is linked to the
overdependence on energy derived from the non-renewable sources, the fossil fuels. The coal-
based thermal power plants 54% of commercial energy produced in India. The environmen-
tal consequences of using coal are serious. The estimates show that the total CO2 emissions
from thermal power plants in the year 2010 is 498,655.78 Gg, the SO2 emissions is estimated
to be about 3,840.44 Gg [3]. The significant portion of commercial energy is spent on the
production of building materials in usable form and their transportation to the construction
sites. The concept of embodied energy must be introduced at the early stage of the construc-
tion and materials syllabi. Table 1 indicates embodied energy contained in some of the con-
ventional building materials.
The environmental consequences of employing the materials must be included as top-
ics of debate and short research assignments. Every process or system of construction
must be presented with environmentally friendly construction systems. Owing to unsus-
tainable consumption patterns and practices, water is becoming an increasingly scarce
resource. In-situ construction system consumes large amounts of water. The water sce-
nario demands a more conservative approach in its use in construction processes. The
introduction of dry construction systems at various levels is imperative. Construction
and demolition waste is posing a new challenge in urban centers across India. Estimated
waste generation during construction is 40–60 kg per m2. Similarly, waste generation dur-
ing renovation and repair work is estimated to be 40–50 kg per m2. The highest contribu-
tion to waste generation comes from the demolition of buildings. Demolition of pucca
(permanent) and semi-pucca buildings, on average generates between 300 kg per m2 and
500 kg per m2 of waste, respectively [4]. The dimension of the C&D waste problem is a
compelling factor to bring in chapters construction waste management built around the
concept of the three R’s. Table 2 Summarizes salient aspects and topics relevant to sus-
tainable construction.
Table 1. Embodied energy in building materials.
Materials Embodied energy in MJ
Burnt brick 3.75–4.5 per unit

Cement 5.85/kg
Lime 5.63/kg
Steel 42.0/kg
Aluminum 236.8/kg
Glass 25.83/kg
Sand (Bangalore) 206/Cu.m.
Stabilized mud block 230 × 190 × 100 × 2.60 2.6/block
Hollow concrete block 400 × 200 × 200 12.30/block
945

Table 2. Summary of salient aspects and topics relevant to sustainable construction.
Aspects of sustainable construction Topics relevant to sustainable construction
Source of materials Finite vs. infinite sources.

Embodied energy Low energy vs. high energy building materials.
Environmental impacts Environmentally friendly construction techniques.
Water conservation Dry construction systems.
The three R concepts Reduce, reuse and recycle concept in construction.
Table 3. Sustainable construction materials and techniques.
Material/construction technique Application
Natural clay, natural clay blended with sand and Substitute with conventional mortar.
lime, quarry dust with lime/cement
Natural soil blocks made of laterite, stabilised Masonry work.
mud blocks, fly-ash walling blocks
Bamboo and agricultural waste Roofing trusses, roofing, walling and flooring
construction.
Dry construction Construction of walls, roofs and floors using
precast and prefabricated components.
Alternative materials Hollow clay roofing and walling, soil-cement
compressed walling and roofing techniques,
ferro-cement concrete, etc.
Recycling of construction and demolition waste Reuse of walling blocks, timber components,
metallic fixtures and concrete waste, etc.
Recycling of waste timber, glass, plastics,
concrete etc to produce new recycled walling,
flooring and roofing materials.
Vernacular Techniques Adobe construction, Dhajji wall construction.
4 FRAMEWORK FOR CURRICULUM IN SUSTAINABLE CONSTRUCTION
The course content in sustainable construction can be carefully structured by identifying the
materials and methods for traditional and contemporary practices. Relevance of indigenous
technologies needs to be re-established and the curriculum should serve as an instrument to
transfer the environmental construction technologies to the field. Table 3 identifies materials
and methods that are deemed sustainable; however the list is only a representative one.
5 NEED TO MOVE BEYOND TOKENISM
The issue of sustainability in construction syllabi needs to be pursued more aggressively. The
course curriculum in sample institutions offer subjects titled as, energy efficient buildings, sus-
tainable architecture/planning, Green architecture/buildings. However, only one such course
of semester duration is made available, in many cases as an elective at higher semesters. These
courses provide only an overview of the subject. The concept and content of sustainability
needs to be entwined across the width and the depth of the entire course curriculum.
6 CONCLUSION
Climate change, rise in sea levels, rise in the earth’s average temperature and ozone layer
depletion are some definite indicators of the extent of the harm that has been caused by
946

human activities. These damages are irreversible and man can only be wise not to worsen the
scenario any further. Construction activities are a major factor in causing adverse environ-
mental consequences. The construction sector needs to be supplied with adequately trained
professionals to implement sustainable construction practice on sites. The obvious responsi-
bility of capacity building in the fields of sustainable construction lies with universities and
part of that responsibility must be shared with architecture institutions, since in most projects
the architects will have the final decision with regard to construction techniques and materi-
als. Construction is fundamentally an act against nature, hence the onus lies with the archi-
tecture and construction community to make this act less damaging. Sustainable architecture
and green buildings are going to be the norms in the construction sector. The university edu-
cation in architecture has a crucial role in capacity building, spreading awareness, promoting
research and development and the transfer of sustainable construction technologies.
REFERENCES
[1] Council of Architecture. (2013). Minimum standards of architectural education–2008. Handbook of

Professional Documents.
[2] Jagadish, K.S, et al. (2009). Alternative building materials and technologies. New Age International
Pvt Ltd. Publishers.
[3] Mittal, Moti L., et al. (2012). Estimates of emissions from coal fired thermal power plants in India.
Retrieved from http://www.epa.gov/ttnchie1/conference/ei20/ session5/ mmittal.pdf.
[4] Waste Management World. (2011). Rebuilding C&D waste recycling efforts in India, 12(5). Retrieved
from http://www.waste-management-world.com. http://www.etymonline.com.
947

Policy and assessment scenario

Effectiveness of housing schemes in rural India: A case study of

Vellanad, Kerala
N. Vijaya, Priyanjali Prabhakaran & A. Lakshmi

Department of Architecture, College of Engineering, Thiruvananthapuram, India
ABSTRACT: The government of Kerala announced a new housing mission in 2016 called

“LIFE”, i.e. Livelihood Inclusion and Financial Empowerment, in accordance with the
national policy of Housing for All 2022. As a preliminary step to the scheme, all incomplete
houses started under previous housing schemes had to be completed. This provided an oppor-
tunity for a detailed study of the incomplete houses. Studying the reasons for these houses
not being completed can prevent such failures in the upcoming housing schemes. A survey
and assessment of these incomplete houses was conducted in Vellanad block panchayat in
the Thiruvananthapuram District. The paper outlines the study and the reasons for the non-
completion of houses under various rural housing schemes in Vellanad block panchayat.
Keywords: housing schemes, effectiveness
1 INTRODUCTION
Rural housing is an indispensable part of community development and village planning.

Rural housing programs are intended to fulfill the housing needs in rural areas, according to
the resources available to the community development agency at the level of the block and
the grama panchayat in the form of finance, technical advice, demonstration, provision of
improved designs and layouts and local building materials and technology. Ideally, such hous-
ing developments should not just provide a roof over their head, but should assist the rural
community in sustaining their livelihoods. Rural housing development usually incorporates
community participation, both physically and financially, in the construction process. This is
evident in the Indian housing policies, which underwent constructive transformations from
the early phases until the recent ones. It played the role of provider in the early phases of
housing development schemes, while it shifted to a facilitator role later, after the decentrali-
zation that was caused by the 73rd Constitutional Amendment Act. Fiscal policies have also
played a role in the shift from free housing to the introduction of schemes, such as the credit
cum subsidy schemes, which involve the participation of the beneficiary. The decentralization
also gave local urban bodies powers and responsibilities, and one such item is housing for the
poor and homeless.
India faced acute housing shortages immediately after independence and, since then,
housing has been a major area of focus of the government as an instrument for poverty
alleviation and for the nation’s development agenda. The early phase of the Five Year Plans,
i.e. from independence until the 1960s, looked into housing migrants and industrial hous-
ing as its main agenda. The phase from the 1960s–1990s saw a change in approach in the
plans, with the establishment of various Housing Boards that had the objective of taking up
housing activities for all sections of society, with a special focus on Lower Income Groups
(LIG). Later, the ninth Five Year Plan declared housing to be a subject for the state, follow-
ing the decentralization caused by the 73rd Constitutional Amendment. The rural housing
program, as an independent program, started with Indira Awaas Yojana (IAY) in January
1996. IAY made a good impact on housing the homeless through its three tier system of
951

releasing installments in three stages, as well as completing the house within a time span
of three years. Although IAY addressed the housing needs in rural areas, certain gaps were
identified during the concurrent evaluations and the performance audit by the Comptroller
and Auditor General (CAG) of India in 2014. Indira Awaas Yojana was followed by Pradhan
Mantri Awaas Yojana, which has now undertaken a rural version known as PMAY (G) with
effect from April 1, 2016.
The homeless statistics in India vary between states, the highest being in Bihar and the
lowest in Daman and Diu (Census of India, 2001). It is of serious concern that the real
cause of homelessness has not really been addressed by our housing experts. Our national
policies have addressed the issues of homelessness, but only on a broad perspective. Planners
and housing experts are not participating in the stage where suitable land parcels are identi-
fied. Land unsuitable for construction, in terms of access to transportation, topography and
infrastructure, is being selected for housing. Construction on such land escalates the cost of
building the house due to high transportation prices. The fund then becomes insufficient to
complete each stage. Beneficiaries of these housing schemes are always from the economi-
cally weaker sections and low income groups. They are forced to contribute not only physi-
cally but also financially to complete the house construction, but fail in most of the cases.
This is due to the mismanagement of factors such as the elderly age of the beneficiary, lack
of motivation and lack of support to start a livelihood to generate an additional income.
2 HOUSING SCHEMES IN KERALA AND THEIR STATUS
State intervention in the housing sector began in Kerala during the 1950s. Many innova-
tive housing programs were developed in congruence with the national policies, raising hope
among the homeless poor of becoming house owners. A housing boom began in the state
in the mid-seventies. Public housing schemes have also had an impressive record during the
past two decades in terms of investment and physical achievements (Gopikuttan, 2002). As
the gap between need and supply decreased, the inequality in housing conditions widened in
Kerala. The poor have become progressively incapable of self-help and mutual help for solv-
ing their housing problems due to various reasons. Thus they have become dependent on a
supporting agency for the execution of their housing projects. The absence of professional
agencies to take up such roles has created unfortunate situations, such as the mismanagement
of funds, poor access to infrastructure, etc.
In India, 32% of rural households live in kutcha structures, but the figure reduces to 19%
in Kerala (Panchayat Level Statistics, 2011). Indira Awaas Yojana (IAY), with its inception
in 2007, has been the most successful housing scheme in Kerala, decreasing the number of
homeless and also establishing the policy of gender mainstreaming. The LIFE mission is
the newest housing scheme for the economically disadvantaged and homeless population
in Kerala as envisioned by the state government. The LIFE mission survey enlists two lakh
homeless populations in Kerala to benefit in the next five years in order to fulfill the national
policy of Housing for All 2022. The research objective of the paper is to explore the reasons
for the incompleteness of houses in Vellanad and summarize strategies to solve it.
3 METHODOLOGY
As part of the LIFE housing mission of the government of Kerala, it was decided, as a first
step, to complete all of the incomplete houses sanctioned under the earlier public housing
schemes. A socio-economic study was done by 18 postgraduate students of Planning (Hous-
ing) at the College of Engineering Trivandrum, guided by two faculties of the department.
The survey was conducted at the block panchayat area to assess the condition of incom-
plete houses and to examine the reasons behind them. Initially, a pilot assessment of five
houses was conducted and a detailed questionnaire was developed. After this, the question-
naire was modified to include all of the necessary data. The research team conducted a two
952

day complete survey of the incomplete houses in Vellanad block on August 10, 2017. The
researchers were divided into eight groups to cover the eight gram panchayats of the Vellanad
block under the guidance of the village extension officers of the respective gram panchayats.
Tribal houses were excluded from the study due to their distinctive housing and socio-eco-
nomic character and they thus require a different approach.
Vellanad is one of the 12 block panchayats of the Thiruvananthapuram District, Kerala. It
is 17 kilometers distant eastwards from the district headquarters. Vellanad is surrounded by
Nedumangad and Thiruvananthapuram Taluks to the west and, Nemom and Perumkadavila
Taluk to the South, and toward the South. Vellanad was chosen as the model cluster by the
state government of Kerala after the preparation of the Integrated Cluster Action Plan for
socio-economic and infrastructure planning and the initiation of spatial planning. Vellanad
block was formed in the year 1962 (Figure 1) and has had housing schemes for the poor
under various government missions. Currently it has several unfinished houses that were
begun under various housing schemes, such as the EMS Bhavana Padhati (state government
scheme) and Indira Awaas Yojana (IAY) from 2007 to 2015.
The construction of houses had started after identifying eligible beneficiaries under various
schemes, such as the EMS housing scheme and Pradhan Mantri Awaas Yojana (PMAY) hous-
ing. A primary survey was done in the eight grama panchayats in Vellanad block, namely Ary-
anad, Kattakkada, Kuttichal, Poovachal, Tholikkode, Uzhamalakkal, Vellanad and Vithura,
using the questionnaire prepared. Later, a socio-economic analysis of Vellanad block was done
to understand the reasons for the delay. Also, a comparison of various factors linked with the
delay was identified. Various analytical methods were used and inferences arrived upon.
From secondary data analysis, it is evident that rural housing in Vellanad had been show-
ing a backward trend from 2006 until 2010, but took a forward leap during 2010–11. The
recorded number of dwelling houses constructed and completed in Vellanad rural sector was
Figure 1. Regional setting of Vellanad block panchayat.
953

Figure 2. Distribution of incomplete houses Figure 3. Percentage of incomplete houses
under various schemes. under various schemes.
highest in the year 2010–11. The ownership shows a changing pattern, with male owner-
ship decreasing from 70% to 61% and female ownership increasing from 28% to 38% of
the total number of houses completed during the years from 2006 to 2011. The number of
pucca houses in Vellanad shows an increasing rate from 66% in the year 2006–07 to 82%
in the year 2010–11. Kutcha houses increased to 26% during the year 2007–08, but have
decreased to 18% during 2010–11. The primary survey summarizes that the national housing
mission, Indira Awaas Yojana (IAY), has been actively executed in the Vellanad block dur-
ing 2010–11, thus affirming that IAY has been able to effectively improve the rural housing
scenario in Vellanad by completing a high number of pucca houses with female ownership.
The year 2016 saw a change in the strategy of rural housing when the government of Kerala
launched its exclusive housing mission named LIFE (livelihood inclusion financial empower-
ment), which incorporates livelihood inclusivity and financial empowerment as its strategies.
The LIFE mission also includes the landless population and dilapidated houses. This mission
will empower the beneficiaries to manage to have a pucca house and increase their living
conditions by the year 2022, coinciding with the national mission of Housing for All 2022.
4 POST SURVEY ANALYSIS AND INFERENCES: QUANTITY

OF INCOMPLETE HOUSES IN VELLANAD, KERALA
Post survey, the team met and had a group discussion to exchange the observations they had
made regarding their grama panchayat.
Certain common characteristics could be identified regarding the issues and, thus, the
analysis criteria were decided based upon them. Under the IAY scheme, 73% of the incom-
plete houses were allotted. Various analytical graphs were generated to analyze the physical
infrastructure, as well as the socio-economic conditions (Figures 2, 3).
5 ISSUES IDENTIFIED IN HOUSING SCHEMES IN VELLANAD,

LEADING TO INCOMPLETE HOUSES
5.1 Choice of unsuitable land

Plots ranging in size from 3 to 6 cents were purchased, depending upon the financial effi-
ciency of the beneficiary. Those belonging to the Scheduled Caste received financial aid up
to Rs. 75, 000 from the Scheduled Caste Welfare Board. Thus, it is evident from this data
that the site selection happened in remote areas, mostly without vehicular access, because
of the lack of sufficient funds for this. One unique case has been noted in Poovachal grama
panchayat, where 95% of the housing scheme fund disbursed has been used up toward the
construction of a retaining wall, the site being on an extremely sloping terrain.
954

5.2 Lack of professional advice and involvement
The survey noticed that the beneficiaries are identified by the village extension officers and
the plans approved by the block extension officers.
None of these stages involved getting expert advice from a professional, such as an urban
planner or a housing expert. The site selection and preparation of the house plans was also
not being undertaken by experienced professionals, such as architects. This led to plans with
excessive floor areas that did not match the funds available from the mission and meant that
the beneficiaries were not able to complete their houses. In certain cases the plan had not
considered the fact that there was only a single occupant in the house, and the floor area was
higher than that required.
Poovachal has the highest number of incomplete houses with a floor area exceeding that
specified by the government housing schemes (Figure 4). Apparently, IAY does not specify a
maximum limit for the floor area for the houses built under it. One specific case was noted in
Vellanad grama panchayat where the beneficiary was a woman aged 65 years and the hous-
ing plot was located about 700 m inside a rubber plantation, without vehicular access. The
location of the house itself causes issues of safety and security.
5.3 Age and health of the beneficiary

A large number of the beneficiaries are senior citizens. Single women who are senior citizens
have been allotted single occupancy houses, which are located in plots without proper access
(Figure 5).
Families with young children may also find the location inconvenient, where mothers can-
not get any assistance for babysitting or for participation in economic activities. About 63%
of the beneficiaries in Vellanad block are middle-aged.
It was evident from the analysis that health issues have played a vital role in creating
incomplete houses in Vellanad (Figure 6). Two beneficiaries in Vellanad grama panchayat
Figure 4. Distribution of incomplete houses

based on built area exceeding 66 sqm. Figure 5. Age category of the beneficiaries.
Figure 6. Summary of various reasons Figure 7. Distance of current stay.

for incompleteness of houses.
955

were facing serious health issues. They needed external support to proceed with the manage-
ment of their house construction.
5.4 Reduced work participation

Beneficiary participation in house construction is affected in many ways. The majority of
households (50.9%) are staying away from the construction site (Figure 7). This has pre-
vented them from participating in the construction work. There has been a low rate of finan-
cial contribution toward the house construction from the beneficiaries, especially the women.
Most of the women are either not employed or physically unfit to join the construction due to
health reasons. Informal sector financing institutions exist in the block panchayat and most
of the beneficiaries have taken loans from such institutions. This has led to a diversion of
earnings toward the interest payments of such loans rather than toward house construction.
6 CONCLUSION
Vellanad represents a rural area in Kerala. The majority of the population is engaged in agri-
culture or related activities. It exhibits a variety of study options along with an active housing
mission status. Four reasons have been identified that affect housing schemes for the poor in
Kerala: (1) inappropriate plot allocation, (2) lack of expert professional advice, (3) the age
and health of the beneficiary, and (4) reduced beneficiary participation. The identified issues
can be solved through two strategies. A rework in the team leading the housing scheme is
one solution, by including urban planners or housing experts to guide the site selection as
well as in preparing appropriate floor plans. Importance is not given to associated home-
based income generating activities, which are usually undertaken by the poor. The research
leaves scope for further interrogation into aspects such as the scope of livelihood inclusion
and methods to incorporate alternative building materials and technology into similar rural
housing development. It is hoped that such studies are undertaken before formulating hous-
ing schemes.
REFERENCES
Census of India, Concepts and Definitions. (2001). Government of India.

Gopikuttan, G. (2002). Public housing schemes for rural poor in Kerala, A critical study of their
suitability.
Housing and Land Rights Network. (2016). The Human Rights to Adequate Housing and Land in
India. New Delhi.
Integrated Cluster Action Plan, Shyama Prasad Mukherji Rurban Mission. Ministry of Rural
Development.
Kannan, K.P. & Imran. Khan. (2016). Housing condition in Kerala with special focus on rural area and
socially disadvantaged sections. Thiruvananthapuram: Laurie Baker Centre for Habitat Studies.
Panchayat Level Statistics. (2011). Thiruvananthapuram: Government of Kerala.
Thiruvananthapuram: Centre for Development Studies.
956

Planning for solid waste management in Kuttanad wetland

region, Kerala State
A. Sanil Kumar & V. Devadas

Department of Architecture and Planning, Indian Institute of Technology, Roorkee, Uttarakhand, India
ABSTRACT: Solid waste is one of the biggest issues in wetland regions worldwide due
to their high-water content and scarcity of land. The inefficiency in the waste management
often leads to pollution of the environment and subsequent degradation of such regions. The
study examines the problems prevalent in the Kuttanad Wetland Region, which is known as
the ‘Rice Bowl of Kerala State’. It is observed that solid waste and its improper handling have
emerged as one of the biggest challenges in this region which is affected by severe deteriora-
tion in its water sources and pollution of its natural elements. The study is based on primary
and secondary data collected from the region and presents a profile of successful solid waste
management strategies through the case discussion of The Netherlands. The paper recom-
mends strategies for solid waste management in wetland regions which could lead to a plau-
sible sustainable development of the region.
1 INTRODUCTION
Municipal Solid Waste (MSW) is one of the biggest challenges in the development of urban
pockets worldwide. The composition of municipal solid waste varies greatly from municipal-
ity to municipality, and it changes significantly with time (Kumar, et al., 2016). The process
of waste management includes an array of tasks which are focused towards generation, pre-
vention, characterisation, monitoring, treatment, handling, reuse and residual disposition
of solid wastes. As compared to the natural ecosystem and the ecological cycle, waste man-
agement focuses on the waste generated from the man-made activities and processes. This
process faces a lot of challenges in situations like India with increasing population, squalor
and dwindling economic conditions. The system of governance is also a big hurdle consisting
of corrupt practices, rapid unplanned urbanisation and lack of resource management. The
development trend has exerted more pressure on the existing system which requires better
management and technological interventions. The management of waste in a wetland region
brings a set of uncertain factors for effective execution. The scope of the research includes
solid waste disposal strategies pertaining to a wetland region. The study is based on primary
and secondary data collected pertaining to the Kuttanad Wetland Region (KWR) which sur-
rounds the Vembanad Lake.
1.1 Characteristics of solid waste in Kerala State

In the case of Kerala State, the quantum of MSW generation varies between 0.21–0.35 kg/
capita/day in the urban centres and it goes up to 0.5 kg/capita/day in large cities (NEERI,
1996). The average daily per capita generation of solid waste comes to 0.178 kg with a very
high variation from 0.034 kg (Koothuparamba) to 0.707 kg (Thalassery) (CESS, 2001;
Padmalal & Maya, 2002; Varma & Dileepkumar, 2004). In the municipalities located within
the Kuttanad Wetland Region, the total waste generation varies between 5 to 43 tonnes per
day while the per capita waste generation amounts to an average of 240 to 250 grams per
day (NEERI, 1996). The latest report (generated in the year 2012) by the Central Pollution
957

Figure 1. Quantum of municipal solid waste generation in Kerala State. The majority of the waste gen-
erated (48%) is through domestic sources, followed by commercial establishments (11%), hotels & restau-
rants (10%), street sweepings (9%), markets, construction, (6%), institutions (5%), among other sources.
Table 1. Municipal solid waste generation as per the standard norms (based on reports from NEERI,
1996; Varma & Dileep, 2004; SEUF, 2006).
Name of Population MSW generation Per capita generation

S. No municipality (2001) tonnes/day (kg/per person/day)
1 Alappuzha 177079 43 0.24

2 Kottayam 60725 15 0.25
3 Chenganassery 51960 13 0.25
4 Thiruvalla 56828 14 0.25
5 Cherthala 45102 11 0.24
6 Vaikkom 22637 5 0.22
7 Kayamkulam 65299 16 0.25
Total 479630 117 0.24
ontrol Board puts the State average figure at around 249 grams per day per person (CPCB,
C
2012). It is observed that there is no drastic change in the total MSW generation in the region.
The waste sourced from domestic sources forms the majority (48 per cent) of the MSW, fol-
lowed by commercial establishments, hotels and restaurants, street sweepings, with waste
coming from markets, construction and demolition activities, institutions/schools, proving to
be the other important components. In the major cities of the State, around 80 per cent of
the waste is compostable organics enabling high level of recycling in the form of manure or
fuel (KSUDP, 2006 & Varma, A., 2009).
2 KUTTANAD WETLAND REGION
The Kuttanad Wetland Region (KWR) is spread over 1471 square kilometres, and is a unique
site where below sea level paddy farming is conducted in India. It is also recognised as a
Globally Important Agricultural Heritage System (GIAHS) by FAO (Food and Agricul-
tural Organization) of United Nations. The KWR is spread over three districts (Alappuzha,
Kottayam and Pathanamthitta) with the river waters reaching the Vembanad Lake from six
districts. The region consists 78 villages, 25 census towns and seven municipalities.
958

2.1 Existing facilities
The district of Alappuzha has a windrow composting facility with an input of 50 tonnes/day
which caters to the existing population of Alappuzha Municipality. Kayamkulam Munici-
pality has a landfill site at Murikkummoodu which is spread over an area of 0.82 acres
catering to about 12 tonnes/day. There are four vermi-composting plants in the Alap-
puzha Municipality. They have a combined capacity of 10 tonnes/day. Other landfill facili-
ties include Chengannur Municipality (Perunkulam Paadom—capacity of 8 tonnes/day),
Mavelikkara Municipality (capacity of 9 tonnes/day) (TCPO, 2011). The Alappuzha munic-
ipality has been recognized as a zero-waste town owing to its successful implementation of
solid waste management strategies. The municipality set up biogas plants, pipe compost
units in households and aerobic composting units in public places. It also set up surveil-
lance cameras across the city, linked to the police control room, to catch those littering
public places (Philip, 2014). The success of this model has resulted in the village panchayats
surrounding Alappuzha taking up aerobic composting units at various locations along the
National and State highways.
2.2 Case discussion—The Netherlands

The selection of The Netherlands as a case discussion for solid waste management strate-
gies was based on the fact that the region has a running coastline of 451 kilometres which is
similar to the Kerala State with coastline of 580 kilometers. In a similar manner, most of its
landmass is coastal lowland and reclaimed land (polders) and it is prone to flooding. Despite
the threat from the surrounding waters, they are one of the leading nations in their waste
management strategy. Their planning was shaped by scarcity of land, proximity to water and
their concerns for the environment—features common to a wetland region. The Dutch Waste
Management Association (DWMA) interacts with about 50 companies which are involved in
collecting, recycling, processing, composting, incinerating and land filling waste. The waste
disposal model considered landfilling waste as the last objective relying on recycling strate-
gies for various waste categories and taxing each tonne of landfilled waste (Feller, 2014). The
Dutch waste management strategy is backed by a strong legislation brought forth through
the efforts of Ad Lansink (known as Lansink’s Ladder strategy) which imbibes stringent
waste management norms in all aspects of daily life.
Some of the waste treatment innovations such as roads made of recycled plastic are pres-
ently being adopted in Kerala State. The identification of waste into different categories will
lead to its effective utilisation in the wetland region. The segregation and recycling of these
waste categories can be executed through the linking of these items with existing taskforces
such as Kudumbasree units, Green Army, self-help groups in individual villages and town-
ships with opportunities for income generation, among others.
Some of the unique waste management innovations brought forth by The Netherlands
waste treatment agencies which can be adopted in the near future include:
• Underground refuse containers with e-access—General use (records each user and their
waste dumping pattern). This system can be adopted in the newly planned housing town-
ships and large-scale technology parks.
• Styrofoam recycling units—Package industry (recycles 100 per cent of the waste). This can
be adopted with a complete recording of the input and output data pertaining to the seven
municipalities (first phase).
• Waste sorting machines—construction industry (recycles 95–98 per cent of the waste).
There can be an amendment to the existing rules and regulations related to the construc-
tion industry to adopt materials with high recyclability.
• Online reporting of waste accumulation—response time linked to online data. This system
can be adopted for major junctions and intersections in the region in the initial phase.
• Methane extraction—from various landfill sites. This was one of the biggest problems
which led to the change in collection and disposal patterns within the Alappuzha munici-
pality. However, it can be adopted to the various landfill sites with added infrastructure.
959

• Nerada technology—converts waste water reusable for agriculture (1/3rd size of a conven-
tional plant). This would be critical to the regions such as Kuttoor panchayat (near Thiru-
valla) and locations surrounding Alappuzha municipality where waste water is directly
dumped into the existing canals.
2.3 Data collection and analysis

2.3.1 Primary survey
A socio-economic survey was conducted in Kuttanad Wetland region with responses sought
from various households distributed within the location. The information collected was part
of a larger survey format conducted as part of a study on development for an integrated
planning framework for Kuttanad Region. The survey locations included Ramankary, Kai-
nakary, Nedumudi, Champakulam, Thakazhy, Edathua, Thalavady, Vechoor, Thalayazham,
Mannancherry, Thaneermukkom, Karuvatta, Thiruvarppu, Kottayam and Nedumpuram.
The sites selected were based on the premise of having optimum representation from vari-
ous locations along with a consideration for most afflicted regions in the wetland based on
primary observation and secondary data. A total of 441 households (1891 respondents) were
covered based on random sampling method in the duration from June to November 2017 and
their responses collated. They were inquired about questions related to waste disposal meth-
ods, agency through which waste was disposed, frequency of collection or disposal, among
others. The respondents were also inquired about their perception regarding the quality of air,
water and soil conditions within the region. They were also requested to record their remarks
regarding the key problems and their root causes within the study region. The following obser-
vations are suggested as per the primary survey and reconnaissance studies within the region:
• The majority of the residents (75 per cent) of the KWR are burning their waste on site,
which is followed by people who bury their waste on site in plastic bags (18 per cent), fol-
lowed by households who throw their waste at random sites (4 per cent) and families with
waste collection facilities (3 per cent).
• Waste collection, wherever applicable, was being conducted daily or on a weekly basis.
Most of the residents (93 per cent) disposed off their waste on a regular basis using burn-
ing or burial methods. The frequency of burning or burial varied anywhere between twice
a week or once a week in most cases.
• There are no charges or expenditure related to waste disposal incurred by the residents.
• Wastes such as bottles or glasses are discarded in an improper manner but burning plastics
and organic waste is prevalent.
• About 54 per cent of the respondents have rated the air quality of the region as Good,
while 43 per cent have rated it as satisfactory while 3 per cent have termed it as poor.
• Regarding the soil quality, 55 per cent have termed it as satisfactory while 41 per cent rate
it good, while the remaining 4 per cent term it as poor.
• In questions regarding quality of life within the region, a majority (89 per cent) have
termed it as good (34 per cent) and satisfactory (55 per cent) while 11 per cent of the sam-
ple population rates it as poor.
• The respondents were asked about the imminent problems within the region. Based on
the ranking method, the improper handling of waste was rated to the biggest problem,
followed by the salinity in water (2), pollution of water sources (3), scarcity of water for
domestic use (4), and an increase in pests (5), among other issues. The possible causes for
these problems were rated in the following descending order—lack of waste management
(1), increase in illegal waste dumping (2), improper agricultural activities (3), lack of inte-
grated planning (4) and obstructed water channels (5), among other causes.
• The commercial establishments (hotels, restaurants, shops) dump their liquid waste directly
into the natural water channels without any treatment.
2.3.2 Field discussions

The author conducted discussions with block and panchayat members from the selected
regions before commencing on the survey. The field study helped in gathering information
960

from the following region related to persistent problems within the region and the upcom-
ing projects proposed by the members which are to be implemented in the coming five to
ten years. Further, the details regarding the region were discussed with various experts from
fields of urban planning, agriculture and allied sciences, scientific research, along with gov-
ernment body officials including the municipality, Town and Country Planning department,
among others to ascertain the issues related to developmental hurdles and solid waste strate-
gies within the region. These discussions were also conducted with local NGOs and the find-
ings documented in this section.
It was observed that solid waste disposal strategies were restricted to the municipalities
with a majority of the villages and panchayats lacking these methods of proper waste dis-
posal. The quality of life of the residents was severely affected owing to the poor quality of
water and excessive waste being dumped directly into the water sources. In the ecological
system, the region is moving towards a great imbalance with a spike in the number of debris,
weeds and floating plants thriving in the environment leading to severe damage to the water
quality. Since, the water channels are connected to the main river streams flowing through
the region, the existing water channels, which have been converted into drains, in turn, pol-
lute the larger water bodies within the region. The presence of plastic (in all its forms) within
the region was resulting in many of the paddy fields turning into dumping grounds for these
waste items. The empty plots within the regions were also being used as dump sites by resi-
dents. It was suggested by one of the experts that the timely cleaning of canals on a regular/
annual basis could be a major step to the solving of other problems related to the region
(Kumar & Devadas, 2015).
2.4 Critical issues

Based on expert discussions and personal interviews, the following issues were projected as
critical regarding efficient waste management:
• Sensitive nature (ecology) of the site
• Human errors in handling of solid waste
• Existing waste at various locations
• Budget constraints owing to unique geographical conditions
• Scarcity of land
• Excess presence of water
• Lack of recycling practices
• Imminent threat of flooding.
• Toxic nature of some wastes.
• Will of the governing authorities.
• Lack of effective or stringent regulations (Kumar & Devadas, 2015).
3 METHODS OF SOLID WASTE DISPOSAL
A comparison was conducted between the various solid waste disposal methods and their
advantages and disadvantages based on their utility in the study area. The various meth-
ods considered included Landfills, Incineration, Pyrolysis, Deep-well injection and Deep-sea
waste disposal. Based on the analysis and expert inputs, it was observed that landfilling and
incineration were more feasible options for large scale solid waste disposal strategies for the
wetland regions.
4 STRATEGIES FOR WASTE MANAGEMENT
Following the analysis of the existing situation, the following recommendations are for-
warded as sustainable strategies for solid waste management.
961

• Integrated Planning for Waste Management—Solid waste management can be taken up as
a priority in planning realizing its potential as a key component to environmental protec-
tion and income generation.
• Waste management through policy planning—A step in this direction is implemented
through the ‘Swachh Bharat Abhiyan’ (Central scheme) and ‘Suchitwa Mission’, ‘Green
Protocol’ or ‘Haritha Keralam’ (State schemes). The various policies are facilitating liquid
and solid waste management through setting up of stable infrastructure for waste disposal.
This has to be extended further considering the harmful nature of certain waste compo-
nents which would result in higher investment in disposal than manufacturing costs.
• Minimisation of waste generation through organised processes—Waste minimisation is
defined as the reduction, to the extent feasible, of pollutant waste that is generated or
subsequently treated, stored or disposed of. (Polprasert, 1996).
• Waste segregation at source—The segregation of waste helps in planning of the separate
treatment methods for each of the waste categories. The cost of transporting mixed gar-
bage is high and segregation at source can reduce the total cost by 30 per cent. (The Hindu,
2013). Many countries which have planned for waste segregation by marking bins with
different colours as per the waste category.
• Waste management as a community initiative—Waste collection and disposal involves the
contribution of all members of the society. Thus, the planning agencies have to project
waste management as a community initiative to consider its best utility.
• Usage of waste for power generation—In the post-segregation scenario, the ‘segre-
gated’ waste items can be processed and utilised as fuel for powering turbines to generate
electricity.
• Composting strategies—One of the primary strategies for a wetland region is the concept
of composting. Considering wetland regions, these facilities needs to be organised in a
planned manner at locations which are less prone to flooding but are generally accessible.
• Decentralised waste treatment strategies—In wetland regions, it is easier to focus on
micro-waste processing units which can convert the waste into a usable form and make it
less harmful for nature.
• Technological Intervention—Waste segregation and its management can be further rein-
forced through the use of technology enabling users to know various inputs regarding
waste disposal strategies.
5 CONCLUSION
The concept of solid waste management in wetland regions has many facets for considera-
tion. The success of the Alappuzha Municipality in adopting the Zero-Waste strategy reflects
that a series of different techniques can be combined together for effective waste manage-
ment. The future research could focus on sustainable technologies which encourage user
participation in waste management strategies. This research suggests that solid waste man-
agement strategies have to combine strong legislative procedures, modern technology and
participatory approach to achieve plausible solutions.
REFERENCES
CESS. 2001. Carrying capacity based development planning of Greater Kochi Region (GKR), Rep. Centre
for Earth Science Studies, Thiruvananthapuram. p. 269.
CPCB. 2012. Status Report on Municipal Solid Waste Management, Rep. Central Pollution Control
Board. New Delhi. p. 9.
Feller, G. 2014. Dutch Successes [Journal]. - Northbrook: Waste Management World, 2014. - 1: Vol. 11.
KSUDP. 2006. Solid waste management of Kollam, Kochi, Thrissur and Kozhikkode Corporations of
Kerala. Dft. Detailed Project Report. Local Self Government Department, Government of Kerala &
Asian Development Bank.
962

Kumar, S. & Devadas V. 2015. Waste Management in Kuttanad Wetland Region. In: Sustainable &
Smart Cities Conference 2015. Surat: SVNIT, Surat.
Kumar, S; Dhar, H.; Nair, V.V.; Bhattacharyya, J.K.; Vaidya, A.N.; Akolkar, A.B. 2016. Characteriza-
tion of municipal solid waste in high-altitude sub-tropical regions. Environmental Technology. 37:
2627–2637.
NEERI. 1996. Municipal solid waste management in Indian Urban Centres. Rep. National Environmen-
tal Engineering Research Institute. Nagpur.
Padmalal, D.; Narendra B.; Maya K.; K., R. Reghunath; Mini, S.R.; Sreeja, R. & Saji, S. 2002. Munici-
pal solid waste generation and management of Changanasseri, Kottayam and Kannur Municipalities,
Kerala. Rep. Centre for Earth Science Studies, Thiruvananthapuram. CESS PR-02-2002. p. 47.
Philip, S. 2014. ‘Clean Home, Clean City’: Alappuzha municipality shows the way. The Indian Express.
[online] pages. Available at: http://indianexpress.com/article/india/india-others/clean-home-clean-
city-alappuzha-municipality-shows-the-way/ [Accessed October 10, 2017].
Polprasert, C. 1996. Organic Waste Recycling—Technology and Management [Book]. - England: John
Wiley and Sons, Vol. 2nd edition.
Special Correspondent. 2013. Waste segregation at source cheaper, effective: expert. The Hindu .[online].
Available at: http://www.thehindu.com/news/cities/Mangalore/Waste-segregation-at-source-cheaper-
effective-expert/article11613715.ece [Accessed October 10, 2017].
TCPO. 2011. IDDP Alappuzha Draft Plan - Alappuzha: yet to be published.
Varma, A. & Kumar, D. 2004. A handbook on solid waste management. Clean Kerala Mission, Govt. of
Kerala. p. 78.
Varma, A. 2009. Status of Municipal Solid Waste Generation in Kerala and their Characteristics.
[online]. Available at: http://www.seas.columbia.edu/earth/wtert/sofos/Ajaykumar_Status%20of%20
MSW%20Generation%20in%20Kerala%20and%20its%20Characteristics.pdf [Accessed October 1,
2017].
963

Green infrastructure as a tool for urban flood management
Reshma Suresh & C.A. Biju

ABSTRACT: Floods have been a frequent occurrence in cities, be they planned or

unplanned. This can be attributed to the increase in constructed spaces, which causes a
decrease in the permeable land. For a long time, governments implemented gray solutions,
such as drains and levees, to mitigate flood risk. However, with the adverse impacts of floods
growing, a need has arisen for a more integrated approach to urban flood risk management.
Green Infrastructure (GI) solutions have emerged as a key component of this integrated
approach.
Keywords: Urban flood, Greay solutions, Green Infrastructure
1 INTRODUCTION
With the increase in impervious surfaces in urban areas, the storm water runoff is overwhelm-
ing the existing infrastructure, which causes flooding and sewer overflows. Under these con-
ditions, cities are under pressure to find cost-effective, sustainable and socially responsible
solutions to urban flood management. This can be done through Green Infrastructure (GI),
which can complement or augment the present solutions for urban flood management. The
term “green infrastructure,” when used for storm water management, denotes techniques,
such as rain gardens, green roofs, permeable pavements, street trees, and rain barrels, that
infiltrate, evapotranspirate, capture, and reuse storm water onsite. GI allows for both “a
reduction in the amount of water flowing into conventional storm water systems (and thus
a reduction in the need to build or expand these systems) and a reuse of storm water at the
source.”
2 GREEN INFRASTRUCTURE
2.1 Definition
Green infrastructure refers to natural or semi-natural ecosystems that provide water utility
services that complement, augment or replace those provided by gray infrastructure. A GI
framework can be developed on any scale, including multinational, national, regional, local
community or on an individual plot. Have suggested three scales: individual, community and
statewide scales. The framework is applied depending upon the relevant goals of the com-
munity and the benefits to the environment.
2.2 Elements of GI
A wide variety of green infrastructure elements are present and, depending upon the location,
preferences, living standards and the goals which communities require, these elements may
differ. All these GI elements are intended to provide a safe environment and may increase the
economic status of the community.
The GI elements are as follows:
965

2.2.1 Green roofs
These are vegetated roofs that are partially or completely covered with plants. They grow in
3–15 inches of soil, sand, or gravel planted over a waterproof membrane. It can be intensive
(175–220 kg. per sq. ft.) or extensive (33–110 kg. per sq. ft.). It reduces the annual storm
water runoff by 50–60% on average, offers the retention of 90% of volume for storms less
than one inch and at least 30% for larger storms, and increases the lifespan of the roof by 2–3
times. Additional benefits include protection from wind damage and UV rays, and regulating
temperature impacts by as much as 21°C.
2.2.2 Rain gardens

Rain gardens are also known as bio-retention basins. These are designed to absorb rain
water that drains from impervious areas (roofs, parking areas, streets, and lawn areas). These
are shallow depressions that are planted with native flowering plants and grasses. They can
reduce runoff because storm water drains into the soil, leading to a reduction in erosion. It
can be made effective by disconnecting downspouts from homes and commercial buildings
that once directed water into the existing storm water management system. Rain gardens can
reduce 40–70% of runoff.
2.2.3 Porous pavements

Porous pavements are permeable pavements that allow rainwater to infiltrate into the soil
and these can be used as pavements, parking lots and pedestrian pathways instead of con-
ventional pavements. They can be made of pervious concrete, porous asphalt, and permeable
interlocking pavers and can reduce runoff volume by 70 to 90%. They can infiltrate 3 inches
of rainwater from a 1-hour storm and have an average life expectancy of 30 to 35 years.
2.2.4 Vegetated swales

These are linear features that have wide and shallow channels present on side slopes, with
natural vegetation to promote infiltration and reduce the flow velocity of the runoff. These
can be used in parking lots, since storm water picks pollution from vehicles, and these can be
treated before entering the watershed. It can reduce the runoff into drains by 50%.
2.2.5 Wetlands
Wetlands receive and treat storm water that is drained from limited impervious areas. These
are aesthetically pleasing and suitable for small wildlife habitats. These do not require a large
amount of space and can be useful in congested urban areas. It is an effective means of man-
aging the more intense and frequent precipitation events. It helps in reducing peak flows and
reducing the intensity of flood events in urban areas.
2.2.6 Planter boxes

Planter boxes are rain gardens that have vertical walls that close at the bottom and they
absorb runoff from sidewalks, parking lots and streets. They are designed to store water tem-
porarily and slowly infiltrate it into the soil. They help to discharge water from terraces into
the soil, rather than into drains.
Figure 1. Porous pavement.
966

2.2.7 Green parking
These are used in parking lot designs that incorporate green infrastructure in their functional
requirements. This also helps in the management of storm water and reduces the heat island
effect. These can be used in parking lots, cycle tracks and in pedestrian walkways. Imperme-
able land can be minimized and reduce the runoff by 25%.
2.2.8 Green streets

Green streets help in integrating all the green infrastructure elements to store, infiltrate and
evapotranspire storm water. Planter boxes, pervious pavements and bio-swales can be used
in the design of green streets. They help to reduce the storm water runoff to a maximum of
about 70 to 90%.
2.3 Benefits of green infrastructure

GI has a predominant importance in creating sustainable communities and in tackling cli-
mate change. It offers a wide variety of benefits to communities and ecosystems. The removal
of pollution, functioning of ground water recharge, flood control and the management of
storm water are the overall benefits of GI. It provides a much better quality of life, a healthy
environment and an efficient transportation system. GI implementation offers multiple ben-
efits in terms of social, health, environmental, economic and climate aspects. A list of sum-
marized benefits are described below to show its importance in each individual aspect.
2.4 Role of GI in flood management

In urban areas, up to half of the area does not allow rainwater to seep into the ground.
This is due to impervious surfaces, such as roads, roofs, parking areas, sports facilities, etc.
Figure 2. Anatomy of a green street.
Table 1. Benefits of GI.
Improve Reduce heat Improve Improve energy Reduce Flood

Type air quality island effects water quality efficiency pollution control
Porous paving Yes Yes Yes Yes

Bio-swale Yes Yes Yes
Planters Yes Yes Yes
Rain gardens Yes Yes Yes
Wetlands Yes Yes Yes Yes
Green parking Yes Yes Yes Yes
Green roof Yes Yes Yes
Green streets Yes Yes Yes Yes Yes
Source: The Center for Clean Air Policy, February 2011.
967

Table 2. Benefits of GI elements of flood management.
Storm water Average peak

control method flow (% removed)
48″ Soil bio-retention 85

30″ Soil bio-retention 82
Constructed wetlands 81
Porous pavement 79
Green roof 65
Bio-swales 50
Green parking 47
Source: The Center for Clean Air Policy, February 2011.
Obviously, urban areas in more developed economies may have a higher share of built-up
areas with hard surfaces. Since the ground is unable to absorb rainwater, the water flows
directly into rivers, streams, and sewers.
The United States Environmental Protection Agency (EPA) estimates that a typical Ameri-
can city block “generates five times more runoff than a woodland area of the same size, while
only about 15 percent (of rainwater) infiltrates into the ground for groundwater recharge”.
During major rainfall events, this additional runoff overwhelms rivers, streams, and sewers
and causes severe flooding. Additional risks include drought (due to reduced groundwater
recharge and reduced surface water storage) and negative impacts on water quality.
The combination of increasing flood risk, the potential for major human and economic
losses, and the unevenness in the efficacy and costs of gray infrastructure has led to a growing
interest in exploring other approaches.
GI solutions focus on managing wet weather impacts by using natural processes. As part
of an integrated flood risk management framework, they can also deliver environmental,
social, and economic benefits, and can be cost-effective, low in impact, and environmentally
friendly (sustainable). In contrast, traditional approaches, such as levees and dams, focus on
changing the flow of rivers and streams to protect local communities, and use piped drain-
age systems in urban areas to quickly move storm water away from the built environment.
As an unintended consequence, fast drainage of water may result in drought problems, and
drought may reduce the ability of existing green spaces to provide important services, such
as reducing heat stress.
Gray infrastructure solutions remain a key component of flood risk management frame-
works and are necessary in many situations. But GI solutions can be a valuable part of an
integrated approach. GI solutions include, among others, wetlands, bio shields, buffer zones,
green roofing, tree pits, street side swales, porous pavements, and the use of green materials
(wood, bamboo, coconut nets, etc.). These measures not only help to reduce flood impacts
but also produce environmental and health benefits.
3 FLOODING IN INDIA
India is the most flood affected nation in the world after Bangladesh. It accounts for 1/5th
of the global deaths by floods every year and on average 30 million people are evacuated
every year. The area vulnerable to flood is 40 million hectares and the average area affected
by floods is 8 million hectares. Unprecedented floods take place every year in one place or
another. The most vulnerable states of India are Uttar Pradesh, Bihar, Assam, West Bengal,
Gujarat, Orissa, Andhra Pradesh, Madhya Pradesh, Maharashtra, Punjab, and Jammu and
Kashmir. A history of floods can be seen from ancient times. In the independent India, the
first major flood occurred in 1953. After this a series of floods happened every year.
3.1 Cause of urban flooding in India

About 63% of the area of India is made up of urban areas.
968

The area under urban settlement in India has increased from 77,370.50 sq. km in 2001 to
102,220.16 sq. km in 2011, showing that 24,850.00 sq. km of additional land area has been
brought under urban use. The land use changes show that there has been a 74.84% increase
in built-up areas and a 42.8% decrease in open spaces between the years 1990 and 2015, with
a substantial increase in urbanization. The vegetation had decreased by 63% by 2015. In the
past 43 years, the increase in peak runoff and runoff volume is marginally varied by 3.0% and
4.45% and the total flood hazard area has increased by 22.27% in urban areas. It is also seen
that there has been a 35% decrease in wetland areas from 1990 to 2015, which is the main
natural basin for storing the rainwater.
Thus, it can be summarized that the major causes of urban flooding in India are:
• land use changes
• surface sealing due to urbanization (which increases runoff)
• occupation of flood plains and obstruction of flood flows
• urban heat island effect (which has increased the rainfall in and around urban areas)
• sudden release of water from dams located upstream of citizen towns and the failure to
release water from dams resulting in a backwater effect
• the indiscriminate disposal of solid waste into urban water drains and channels.
4 GI IN INDIA
Green infrastructure has been practiced in India for a long time, but not in a co-ordinated
manner. Not all the elements of GI are implemented in a place, which can lead to multiple
and long-term benefits. Many of the elements are being successfully implemented, but are
not using the benefits of their combined use. For example, green roofs are being practiced
effectively in Bangalore, Hyderabad. The rain gardens are implemented in Samshabad air-
port. The rainwater harvesting is also an element of GI that is being widely practiced in
India. Porous pavements, on the other hand, have not been used in India until now, so their
potential for urban flood management has not yet been explored.
5 CASE STUDIES AND ANALYSIS
5.1 Case studies and analysis

There are various examples from across the world where green infrastructure has scored over
the contemporary gray infrastructure. They throw light on the integration of each element of
GI and how successful each project was.
5.2 Case study themes for consideration

It can be difficult to compare GI projects—each is very context specific and the process
through which each comes into function depends on the existing planning system. However,
a number of common themes arise that can be used to make GI projects successful. They are
1) Approach 2) Regulatory framework 3) Implementation and monitoring 4) Results.
5.3 Case studies and analysis: Methodology

The case studies of GI across various projects were considered and categorized. The general
details of the city where GI was initiated, origin of intervention, approaches, effects and its
role in urban flood management were considered for each of the projects.
5.3.1 Portland, Oregon

Portland is the biggest city in the northwestern state of Oregon. It has a population of over
600,000, and is highly urbanized. The city has a history of major floods. Portland has 50%
impermeable land surfaces (60% attributed to streets, 40% to rooftops). In 2004, Portland
969

experienced 50 overflow events. The city has failed to deal with it by the construction of
dams, flood walls, etc. The city is a prime example of green storm water management.
Approach: The city adopted a comprehensive multifaceted approach, which includes regu-
lations, incentives and the monitoring of results. In 2007 a green streets program was started
to incorporate the use of green street facilities in public and private developments.
Regulatory framework: A storm water management manual and code, which outline the
requirements. In order to develop or redevelop over 500 ft2 of impervious surface, the city
requires pollution reduction and runoff control standards.
Implementation and monitoring: To monitor progress and compile good practices, a floor
area bonus was provided for green roofs, which incentivized residents to create an eco roof in
exchange for an increase in a building’s allowable area.
Results: GI of Oregon is a mature and comprehensive GI program. By 2014, 330,000 ft2
of eco roof installations came into existence. Green street projects retain and infiltrate about
43 million gallons per year, with the potential to manage 40% of Portland’s runoff annually.
Portland invested 52 crores in green infrastructure to save 1,621 crores in hard infrastructure.
5.3.2 Milwaukee, U.S.

Milwaukee is the largest city in the state of Wisconsin and the fifth largest city in the Mid-
western United States. The Milwaukee region has steadily grown since the year 2000. As the
region continues to grow, so does its impervious surfaces. The more housing, pavements and
concrete increase, the more absorbent wetlands diminish. The region has continuously faced
threats of flash flooding, causing damage to buildings and vehicles. A minimum of 10 floods
in a year were faced by the city.
Approach: Focused on downspout disconnections and rain barrels. Sixty rain gardens were
installed to control runoff, with twenty thousand square foot of green roof on a local housing
project that will retain 85% of runoff, with the remaining 15% redirected to rain gardens and
retention basins for onsite irrigation.
Regulatory framework: Several programs and initiatives were carried out. “Every Drop
Counts”- shows residents how to reduce storm water by installing rain barrels and planting
rain gardens. In co-operation with the Regional Planning Commission, the sewerage district
also implemented the green streets program, which involved the purchase and protection of
undeveloped lands and open spaces adjacent to streams, shorelines, and wetlands. In 2010, the
Milwaukee Metropolitan Sewerage District awarded 24 crores in green infrastructure grants
to 14 groups. Through the grants, 7,500 square feet of permeable pavements, 4,000 square
feet of green roofs, 1,100 square feet of bio-swales and rain gardens, and two 1,000 gallon
rain harvesters were implemented.
Results: GI in Milwaukee provides a 31 to 37% reduction in storm water flow, a 5 to 36%
reduction in peak flows, and a 14 to 38% reduction in Combined Sewer Overflow volume.
Figure 3. Existing green infrastructure types.
970

5.3.3 Toledo, Spain
Toledo is a municipality located in central Spain. It is about 232.1 km2 in area with a popula-
tion of 83,459 in 2016. The entire city continuously faces flooding, even during small amounts
of rainfall, which affects the traffic movement. In January 2005, some of the major roads were
closed for 24 hrs due to flooding. Since 2006, the use of policies and demonstration projects to
help promote green infrastructure has drastically reduced Combined Sewer Overflow.
Approach: Institutionalizing green infrastructure through demonstration and restoration
projects. New storm water fee systems by using green parking, green roofs and pervious pave-
ments, with the goal of reducing combined sewer overflows while maximizing the net present
value of benefits.
Regulatory framework: Stringent storm water regulations for all new construction and rede-
velopment. A number of ordinances to accelerate green infrastructure investments and water
billing fees will be determined by calculating the amount of impervious cover on a given prop-
erty. Financial incentives to retrofit properties with green infrastructure were also provided.
Results: There are 280 crores in benefits, compared to 778 crores for the gray infrastructure
option. A 60% reduction in CSO volume and a 25% reduction in storm water flow.
6 INFERENCE
From the studies it can be inferred that:

• Indian floods are mainly due to the increase in urbanization, with related factors such as
land use conversion, decrease in open spaces, decrease in permeable land, etc.
• From the case studies it can be inferred that about 30–40% of the urban floods can be
reduced by incorporating green infrastructure practices
• Also, the context in India is comparable with that of the cities where green infrastructure
is applied
• With effective recommendations, urban floods in India can be tackled to an extent
7 RECOMMENDATIONS FOR INCORPORATING GI

FOR URBAN FLOOD MANAGEMENT
The development pattern in India shows an increasing trend toward urbanization and there
has been a major investment in gray infrastructure. Green infrastructure has not been co-
ordinated or integrated at any levels, even though awareness about sustainable development
is on the rise. There is no special policy or strategy for Indian cities. The various elements of
GI are fragmented and the potential is not fully utilized.
1. A limit to the amount of impervious space must be incorporated in guidelines based on
the context.
2. To make it more successful, incentives and subsidies must be given to those who are
involved in such practices.
3. The cities can be categorized in the order of flood severity and the elements of GI can be
incorporated accordingly, based on the context.
4. Find opportunities in existing regulations: examine whether/how current permits and bye-
laws can cover the new activities and create awareness of the same.
5. The approach must only be limited to open spaces, roads, apartments and houses. It must
also be implemented in government buildings.
6. A town planning scheme can be the method through which GI can be integrated and can
act as a platform to improve the resources of GI.
8 CONCLUSION
Over the last two decades, urban planning orthodoxy has promoted a compact urban form
and higher densities to reduce energy consumption and the ecological footprint of cities.
971

However, as outline, densification efforts often pose problems for urban drainage systems,
while brownfield sites that are targeted for development may actually serve more important
functions in terms of water retention, recreational uses and urban cooling.
This paper focuses on the GI approaches as a means to manage urban flood risk. The
paper does not oppose the use of traditional gray infrastructures, as they can be the most
appropriate solution in certain circumstances. However, the paper challenges the dominance
of gray solutions, such as levees and dam flood walls, as a solution for flood management
rather than sustainable green solutions.
ACKNOWLEDGMENT
The authors gratefully acknowledge the contributions of faculty, Government Engineering

College, Thrissur, friends and family for successful completion of the study.
REFERENCES
Chatburn, & Craig (2010). Green Infrastructure Specialist, Seattle Public Utilities. Interview by Sarah
Hammitt.
Hoang, L. & Fenner, R.A. (2015). System interactions of storm water management using sustainable
urban drainage systems and green infrastructure. Urban Water Journal, 2016, 1–21.
Lennon, M. & Scott, M. (2014). Urban design and adapting to flood risk: The role of green infrastruc-
ture. Journal of Urban Design, 19(5), 745–758.
Opperman, J.J. (2014). A flood of benefits: Using green infrastructure to reduce flood risks. Arlington,
Virginia: The Nature Conservancy.
Soz, S.A. & Kryspin-Watson, J. (2016). The role of green infrastructure solutions in urban flood risk
management. Urban Flood Community of Practice, 25(3), 12–20.
Valentine, L. (2007). Managing urban storm water with green infrastructure: Case studies of five U.S. local
governments.
972

Accessibility and interactions with urban blue spaces:

A case of Conolly Canal
Shahana Usman Abdulla & C.A. Bindu

ABSTRACT: Urban waterways represent a potential site for interaction with nature in a
busy urban environment, what we here refer to as “blue spaces”. Blue spaces need to be
viewed as amenities and given importance as green spaces. Few research has been done on
the accessibility of and interaction with blue spaces in an urban area. Through this paper, we
are trying to study how a blue space in an urban area is perceived by the nearby residents and
the reasons behind such a perception, whether positive or negative. A case of Conolly Canal
passing through Kozhikode city is studied and analyzed to find whether it is perceived as a
positive or a negative amenity and what factors lead to such a social impact.
1 INTRODUCTION
The study of different aspects of urban planning are often more complex as they need to take
a number of details into consideration. Normally, attention is given to the physical, social
and economic environment. Social environment has to be given major priority as all the
interventions are ultimately meant for social welfare. Thus research has to be focused more
on how people perceive different elements of physical environment in order to direct plan-
ning interventions in the right way.
There is a growing literature showing how proximity to urban green space can produce
improved health outcomes like reductions in obesity, diabetes and cardiovascular morbidity
(Cutts, Darby, Boone, & Brewis 2009; Ngom, Gosselin, Blais, & Rochette 2016). Urban green
spaces are not limited to terrestrial parks and open areas, but also include urban waterways.
The benefits provided by water features have been widely acknowledged, both as ecological
services (e.g., carbon sequestration, oxygen production, noise reduction, microclimates, etc.)
and as places that are used for recreation and social interaction (e.g., exercise, sport, etc.)
(Kumar 2010, Kondolf & Pinto 2016). In this paper, we are trying to explore how local resi-
dents experience an urban blue space in a sample of neighborhoods in Kozhikode.
2 BLUE SPACES
As blue spaces, we consider hydrographic features that can be waterbodies (e.g., estuaries, ice
masses, lakes and ponds, playas, reservoirs, and swamps and marshes) or flowlines that make
up a linear surface water drainage network (e.g., canals and ditches, coastlines, streams and
rivers) (USGS 2015).
In the sparse blue space literature that does exist, coastal waterways were shown to provide
quality of life benefits, and residents most frequently visited waterways closest to where they
lived (Cox, Johnstone & Robinson 2006). Another study explored distance to stormwater
ponds in Florida, finding that economically stressed census block groups in the inner-city
community tended to be located closer to stormwater ponds with less quality, diversity, and
size (Wendel, Downs, & Mihelcic 2011). Meanwhile, inland urban waterways such as riv-
ers and canals remain understudied as neighborhood amenities with potential impacts on
973

Figure 1. Conolly Canal in Kozhikode town.
urban households. Two meta-analyses focusing on the impacts of blue space on mental health
(Gascon et al. 2015) or long-term human health (Völker & Kistemann 2011) found inad-
equate evidence due to the limited amount of empirical research on the topic.
The factors like accessibility of households to the blue space, whether they interact with or
use the blue space, purpose of visit, time spent and influence on daily lives are being studied
in this paper. A similar study in Northern Utah, United States, showed the blue space they
had considered as a positive amenity based on social perception and accessibility.
3 STUDY AREA
Conolly Canal, usually called as Canoly Canal, is the part of the West coast canal network of
Kerala and runs through the Kozhikode city. It was constructed in the year 1848 under the
orders of the then collector of Malabar, H.V. Conolly.
The canal stretching through Kozhikode town is about 11.4 km long and connects
Akalapuzha in the north and Kallai puzha in the south of Kozhikode town. The width of canal
varies between 6 to 20 m and the water depth during the monsoon ranges between 1 to 3 m.
4 METHODOLOGY
The survey was conducted by a team of 8 members, out of which 3 were engineers of
Kerala State Pollution Control Board, an official from Town Planning Department and 4
post graduate students.
The study comprised of a socio economic survey covering the residents and stakeholders
living along the banks of the stretch. The stakeholders involved included not just households,
but commercial establishments, hospitals and industries. The different types of interaction
with the blue space by the stakeholders and the factors responsible were enquired through
the survey. This paper concentrates on the social perception of the urban blue space and the
factors responsible for such a perception, positive or negative.
The canal portion to be surveyed was divided into 8 stretches, each stretch being approxi-
mately 2 km. The stretches are in between the following points as shown in Figure 2.
1. Eranjikkal
2. Kunduparamba
3. Modappattupalam
4. Ashirwad lawns
5. Sarovaram Biopark
6. Kalluthamkadavu
974

Figure 2. Stretch of Conolly Canal.
7. Mooriyad
8. Kallai
9. Kothi bridge
There are 97 industries on the banks of the canal and river stretch of which 54 were wood
based units. Also there are 501 residences and 208 commercial establishments and seven hos-
pitals. 835 stakeholders including residents and businesses located on the banks of the water-
way stretching from Elathur to Kothi estuary, were taken for the survey.
The survey characterized households on the basis of having direct access to the canal.
Respondents were asked whether the canal affects their lives or not and if yes, how? They
were asked if they use the canal so as to analyse the purpose for which it was used. The
analysis also explores why the canal is not being used or accessed. Finally, the involvement of
people in the revival of the canal is assessed.
5 RESULTS
5.1 Do you have direct access to canal?

Majority of the nearby households are not having spatial access to the canal. The households
nearby the waterway having direct access are said to experience the impacts of the blue space
and themselves have an influence on the water body.
975

5.2 Does the canal affect your life in any way?
The majority of the respondents said that they lives are affected by the canal in one way
or the other. While some said they are unaffected by it. The reasons mentioned for being
affected by the canal were:
• Foul smell due to wastes dumped
• Flooding during rainy seasons
• Mosquito breeding
• Stagnant water with foul smell
• Disinterest in accessing a polluted water body
• Mixing of canal water in the well
5.3 Do you use the canal?

Majority of the respondents doesn’t use the canal for any purposes. The few people who use
the canal used it for the following purposes:
• Seasoning of wood
• Dumping domestic waste
• While using the bridge
• Drainage
The majority does not use the canal stating various reasons, the major one being pollution
of water creating an unhealthy environment. Conolly Canal has a history of people engaged
in fishing, bathing and swimming in the water. But these activities have ceased now due to the
unhealthy condition of the water.
5.4 Do you use canal for waste disposal?

Even though majority have claimed not disposing wastes in the canal, few have admitted
disposal of the following types of wastes:
• Wastes from wood industries
• Domestic wastes
• Septic discharge
• Sewage
Figure 3. Access to canal. Figure 4. Response to influence of canal on lives.
Figure 5. Response to use of canal. Figure 6. Response to waste disposal.
976

Figure 7. Response to revival.
• Drainage
• Industrial wastes
• Hospital wastes
5.5 Major causes of pollution

The major causes of pollution of water in canal are said to be due to dumping of plastic
wastes, industrial wastes, hospital wastes, slaughter waste, market waste, domestic waste,
stagnant water, septic discharge, sewage disposal, inefficiency of waste management and lack
of awareness about importance of canal.
5.6 Would you cooperate with any revival projects for Conolly Canal?
There are damages in the lining provided along the canal and this was majorly attributed to
the encroachment by properties along the stretch of canal. This kind of an illegal intrusion
might create an aversion among the people involved to refrain from cooperating with any
revival projects.
The positive response shows the concern and will of the people to retrieve health of the
urban ecosystem. The very few who restrain from opting to cooperate with any measures to
revive the canal say so as they fear losing their land through land acquisition and due to the
fear of getting caught of illegal encroachment of canal area.
6 CONCLUSIONS
The results show that the urban blue space that we have taken is perceived as a negative amen-
ity by the people of the neighborhood. The major reason for such a result is attributed to the
increase in water pollution due to improper management. People seem to be disinterested in
accessing or spending time at the canal. Though, a blue space has multiple ecological, social
and recreational benefits, the canal under study turned out to have a negative influence on
the lives of urban residents.
Similar study done in Northern Utah indicated the perception of the urban blue space
studied, as a positive amenity. This shows that an ecologically healthy or restored waterway
with public access opportunities can contribute to an aesthetically pleasing experience. On
the other hand, unmonitored or poorly managed urban waterways can be sites of flooding
risk, insect pests, pollution and/or waste disposal. Finally, even ecologically sound wetland
systems can be perceived by humans as disamenities, due to the smell of anaerobic decompo-
sition and the insect populations that thrive in them.
A blue space is an important asset for an urban environment. Planning interventions can
turn any negative influence a blue space has into positive. Planners need to promote use of
and familiarity with urban waterways in order to maximize benefits to local residents and
communities. Restoration is increasingly advocated as a strategy facilitating public access
and use of urban waterways. Infrastructure development like provision of towpaths, naviga-
tional aids, fencing and lighting can improve accessibility and safety. The water quality can
be improved through solutions like swinging weed gates, air curtains, physical removal, flush-
977

ing channels and waste management. Regulations and awareness is required for the proper
conduct near the waterway. Any development along the waterway should follow the pillars of
sustainability, namely the economic, environmental and social justice.
7 LIMITATIONS
The results of the study may or may not be generalizable to other regions. It will depend on
the social structure, built environment and trajectories of urban growth of the region under
consideration. The study only covers aspects of social perception of urban blue space. Also,
the study is limited to canal in an urban area. The results might change with difference in the
type of urban blue space.
8 FUTURE RESEARCH
Further research can be done considering more aspects of social, ecological and recreational
aspects. Interventions necessary to revive and elevate the potential of a blue space in an
urban area can be studied.
ACKNOWLEDGEMENT
The authors are immensely grateful to Mr. K. V. Abdul Malik, Regional Town Planner,
Kozhikode for granting access to data concerning Conolly Canal. We acknowledge the con-
tributions of faculty, Government Engineering College, Thrissur, friends and family for the
successful completion of the study.
REFERENCES
Cox, M.E., Johnstone, R., & Robinson, J. 2006. Relationships between perceived coastal waterway con-
dition and social aspects of quality of life. Ecology and Society 11(1): art35.
Gascon, M., Triguero-Mas, M., Martínez, D., Dadvand, P., Forns, J., Plasència, A., et al. 2015. Mental
health benefits of long-term exposure to residential green and blue spaces: A systematic review. Inter-
national Journal of Environmental Research and Public Health 12(4): 4354.
Historic Alleys, Historic Musings from a Malabar Perspective. 2017. Retrieved from: http://historical-
leys.blogspot.in/2017/07/conolly-and-calicut-canal.html.
Kondolf, G.M., & Pinto, P.J. 2016. The social connectivity of urban rivers. Geomorphology 277: 182–196.
Kumar, P. 2010. The economics of ecosystems and biodiversity: Ecological and economic foundations.
La Rosa, Daniele. 2014. Accessibility to greenspaces: GIS based indicators for sustainable planning in a
dense urban context. Ecological Indicators 42: 122–134.
Melissa Haeffnera, Douglas Jackson-Smith, Martin Buchert, Jordan Risley. 2017. Accessing blue
spaces: Social and geographic factors structuring familiarity with, use of, and appreciation of urban
waterways. Landscape and Urban Planning 167: 136–146.
USGS (2015). National hydrography dataset. Retrieved from: http://Nhd.usgs.gov.
Völker, S., & Kistemann, T. 2011. The impact of blue space on human health and wellbeing—Salutoge-
netic health effects of inland surface waters: A review. International Journal of Hygiene and Environ-
mental Health 214(6): 449–460.
Wendel, H.E.W., Downs, J.A., & Mihelcic, J.R. 2011. Assessing equitable access to urban green space:
The role of engineered water infrastructure. Environmental Science & Technology 45(16): 6728–6734.
978

Energy efficiency governance in an Indian context
V.P. Shalimol
Urban Planning, Government Engineering College, Thrissur, Kerala, India
K.M. Sujith
ABSTRACT: The sensible consumption of energy plays a major role in providing sustainable
development and this responsibility comes through good governance and practice. Governance
has been known to India from past millennia through Kautilya. In the present world, account-
ability, transparency, inclusiveness, equitability, etc. are the key ingredients of good govern-
ance. Thus, energy efficiency, sustainability and governance are interconnected. These concepts
have brought together carbon emissions, climate change, adaptation and mitigation, as well as
employment and poverty reduction. The concept of energy efficiency interlinks these thoughts.
It involves legislative frameworks, funding mechanisms and institutional arrangements, which
go together to support the implementation of Energy Efficiency (EE) strategies, policies and
programs. The government, EE stakeholders and the private sector should work together to
achieve this. However, India’s population makes up 18% of the world’s population, and its
energy consumption is 6% of the world’s primary energy use, which makes it one-third of the
global average. Energy consumption is always on the rise. Energy efficiency and its influence on
the governance sector is analyzed through this paper, which includes the laws and decrees, strat-
egies and action plans, funding mechanisms, implementing agencies, internal assistance, etc.
1 INTRODUCTION
The dimensions concerned with governance include the possession of puissance, the com-
petency to make decisions, how the people’s voices are perceived aurally and how accounts
are rendered. The qualities of good governance were long ago expounded by Kautilya in his
treatise Arthashastra as follows: “In the jubilance of his subject lies his jubilance, in their
welfare his welfare, whatever please himself he shall not consider good”.
The twelfth five year plan (2012–2017) defines good governance as an essential aspect in
order for society to be well functioning. It provides legitimacy to the system by providing
citizens with a way of effectively using resources and by the deliverance of services. The key
ingredients of good governance include accountability, transparency, inclusiveness, equita-
bility, sustainable development, etc. Good governance has always played a critical role in
advancing sustainable development. Thus, good governance and sustainability are the two
faces of a coin. For the sustainable emancipation of institutions and countries, energy effi-
ciency undertakes major functions. Thus, energy efficiency, sustainability and governance
can be viewed in line with carbon reduction, climate change, adaptation and mitigation, as
well as employment and poverty reduction. The way by which energy efficiency can be incor-
porated with sustainable development is by energy efficiency governance.
2 ENERGY EFFICIENCY GOVERNANCE
The International Energy Agency (IEA), with financial support from the European Bank for
Reconstruction and Development (EBRD) and the Inter-American Development Bank (IDB),
conducted a study on energy efficiency governance. Energy Efficiency (EE) governance includes
979

a combination of legislative frameworks and funding mechanisms, institutional arrangements
and co-ordination mechanisms, which will fortify the implementation of energy efficiency strate-
gies, policies and programs by collaboration. From individual households to immensely colossal
factories, EE governance cumulates technology development, market mechanisms and regime
policies. In order to achieve sustainable economic development, the regime, EE stakeholders and
the private sector must collaborate at the required scale and timing of energy efficiency amelio-
rations. The EE governance policy landscape and its understanding help to develop its efficacy.
Various policies and strategies have been followed throughout the world. They include
energy efficiency codes, labels, and incentives. Some examples of this are:
Russia
Code
• Thermal performance of buildings
Labels
• Energy efficiency class of multifamily buildings
• Green standards (2010)
Canada
Code
• Alberta building code (2011)
Labels
• BOMA Best (Building Environmental Standards) Version 2
• ENERGY STAR Portfolio Manager Benchmarking Tool
• LEED Canada (2009)
• LEED Canada (Existing Building: Operations & Maintenance)
Incentives
• EcoENERGY Retrofit (2007)
2.1 Drivers of the energy efficiency policy

There are countrywide variations in the context of energy efficiency, and the typical drivers
are broadly categorized below:
• Energy security
• Economic development and competitiveness
• Climate change
• Public health
2.2 Barriers to energy efficiency

The factors influencing energy efficiency improvement include market, financial, institutional
and technical barriers. Every country has these barriers. In order to overcome these barri-
ers each country will have their energy efficiency policies. The barriers to energy efficiency
include: market barriers, financial barriers, information and awareness barriers, regulatory
and institutional barriers, and technical barriers.
2.3 Policies used to address these barriers

Pricing mechanisms
The introduction of variable tariffs where higher consumption levels invoke higher unit prices.
Regulatory and control mechanisms

Energy audits and energy management, Minimum Energy Performance Standards (MEPS),
energy consumption reduction targets, and EE investment obligations on private companies.
980

Figure 1. Framework of energy efficiency governance.
Fiscal measures and tax incentives

Grants, subsidies and tax incentives for energy efficiency investments. Direct procurement of
EE goods and services.
Promotional and market transformation mechanisms

Public information campaigns and promotions, inclusion of energy efficiency in the school
curricula, appliance labeling and building certification.
Technology development
This includes the development and demonstration of EE technologies.
Commercial development and capacity building

This involves the creation of Energy Service Companies (ESCOs), training programs, and the
development of the EE industry.
Funding remediation
The introduction of revolving funds for EE investments, project preparation facilities and
contingent financing facilities come under this heading.
2.4 Framework of Energy Efficiency (EE) governance

Enabling frameworks, institutional arrangements and co-ordination mechanisms are the
three main governance areas (Energy efficiency governance, Handbook).
3 ENERGY EFFICIENCY GOVERNANCE IN INDIA
India is home to 18% of the world’s population and its primary energy use is 6% of the world’s
consumption, therefore the energy per capita consumption is only one-third of the global aver-
age. India has been responsible for virtually 10% of the incrementation in global energy demand
since 2000. Its energy demand in this period has virtually doubled, pushing the country’s share
of global demand up to 5.7% in 2013 from 4.4% at the commencement of the century. As the
country is progressing, with rising incomes and a better quality of life, there will be a greater
demand for energy. Coal now accounts for 44% of the primary energy mix. Oil consumption in
2014 stood at 3.8 million barrels per day (mb/d), 40% of which is utilized in the transportation
sector. Demand for diesel has been particularly strong, now accounting for some 70% of road
transport fuel use. This is due to the high quota of road freight traffic, which tends to be diesel-
powered, in the total usage and additionally to regime subsidies that kept the price of diesel
relatively low (this diesel subsidy was removed at the cessation of 2014; gasoline prices were
deregulated in 2010). On both the supply and demand sides, India is trying to meet its demand.
LPG use has increased rapidly since 2000, reaching over 0.5 mb/d in 2013 (LPG is second only
981

to diesel among the oil products, pushing gasoline down into an unwontedly low third place).
Nevertheless, the elevation in LPG consumption reflects the growing urbanization, as well as
perpetuated subsidies. Natural gas makes up a relatively minute portion of the energy mix (6%
in 2013, compared with 21% globally). It is utilized mainly for power generation and as a feed-
stock and fuel for the engenderment of fertilizers, however, it nevertheless has a diminutive
but growing role in the residential sector and as a conveyance fuel. Hydropower, nuclear and
modern renewables (solar, wind and geothermal) are utilized predominantly in the potency sec-
tor, but play a relatively minute role in the total energy mix. This is done by encouraging invest-
ments in renewable energy on the supply side and increasing energy efficiency and conservation
on the demand side. The reduction of carbon emissions plays a significant role. There are four
basic standard policy measures for reducing carbon emissions:
• Energy efficiency improvement measures
• Command and control measures
• Domestic carbon taxes
• International emission trading regime of the kind envisaged for the Annex B countries.
The most desirable among these are the energy efficiency measures. The energy efficiency
of the governance of India is analyzed under the following headings.
3.1 Energy efficiency laws and decrees in India

Energy Conservation Act, 2001
This act was enacted in October 2001, to provide for the efficient utilization of energy, its
conservation and matters connected therewith. The Bureau of Energy Efficiency (BEE) was
established with effect from March 1, 2002. Its mission is to develop policies and strategies
with a thrust on self-regulation and market principles, within the overall framework of the
act. The prime objective is to reduce the energy intensity of the Indian economy.
The Electricity Act, 2003
This act was enacted to harmonize and rationalize the provisions of existing laws and to reform
legislation by the promotion of efficient and environmentally benign policies; the act mandates
efficiency in various forms in generation, transmission and distribution. Under the provisions
of section 3(1) of the act, the central government brought out the National Electricity Policy for
the development of the country’s power system, based on the optimal utilization of resources.
3.2 Strategies and action plans in India

The main objective of the eleventh plan was to reduce the energy intensity per unit of Green-
house Gas (GHG) by 20% from the period 2007–08 to 2016–17. When GOI launched the
National Action Plan for Climate Change (NAPCC), this formally addressed India’s objec-
tive for GHG emission reduction. The NAPCC relies on eight missions, of which the National
Mission for Enhanced Energy Efficiency (NMEEE) is a critical one.
Eight national missions

The eight national missions that form the core of the National Action Plan are;
• National Solar Mission
• National Mission for Enhanced Energy Efficiency
• National Mission on Sustainable Habitat
• National Water Mission
• National Mission for Sustaining the Himalayan Ecosystem
• National Mission for Green India
• National Mission for Sustainable Agriculture
• National Mission on Strategic Knowledge for Climate Change.
3.3 Funding mechanisms in India

A steady and reliable source of funding is essential for energy efficiency institution programs.
This can be co-ordinated with the budget allocation, which is undertaken annually.
982

Venture Capital Fund for Energy Efficiency (VCFEE)
BEE has institutionalized VCFEE in India to encourage equity investment in EE projects.
The fund shall provide last mile equity support to specific energy efficiency projects, lim-
ited to a maximum of 15% of the total equity required, through special purpose vehicles or
Rs. 2 Cr, whichever is less. The support has only been provided to government buildings,
private buildings and municipalities.
ICEEP (India Chiller Energy Efficiency Project)

In association with the Ministry of Environment and Forest (MoEF), GOI and IDBI Bank
Ltd by World Bank, ICEEP was implemented from August 2009. The phasing out of use of
Chlorofluorocarbon (CFC), under the Montreal Protocol, and the achievement of energy
efficiency in the refrigeration and air conditioning sector is the objective of ICEEP.
3.4 Implementing agencies in India for EE

The Ministry of New and Renewable Energy (MNRE), the Ministry of Environment and
Forests (MoEF), the Ministry of Power (MoP) and the Ministry of Urban Development
are the implementing agencies that introduced sustainability components with overarching
policy objectives to promote energy conservation in buildings.
3.5 International assistance in India for EE

Partnership to Advance Clean Energy (PACE)
The Partnership to Advance Clean Energy (PACE) is the lead program on clean vitality
between the U.S. and India to mutually chip away at the scope of issues related to energy
security, clean energy and environmental change. PACE looks to quicken comprehensive,
low-carbon development by supporting exploration and the sending of clean energy inno-
vations and strategies. PACE consolidates the endeavors of a few government and non-gov-
ernment partners on both the U.S. and Indian sides and incorporates three key segments:
Research (PACE-R), Deployment (PACE-D), and Off-Grid Energy Access (PEACE).
PACE includes a USD 20 million, five year specialized help (TA) program, which is driven
by the U.S. Organization for International Development (USAID) and the U.S. Bureau of
State and is executed in association with the Ministry of Power (MoP) and the Ministry of
New and Renewable Energy (MNRE). The program partners for this include state, public
and private sector agencies. Their energy efficiency initiatives include:
• Support for the development of the Smart Grid Regulatory framework was provided
• Assisting BEE in the technical update of the Energy Conservation Building Code
• Supporting the Nalanda University and Uttar Haryana Bijili Vitran Nigam headquarter
to became zero energy user
• Supporting the development of a waste heat utilization policy
• Promoting market transformation activities for heating, ventilation and air conditioning.
International Solar Alliance (ISA)

The Alliance, India’s brainchild to bring together 121 solar-rich countries on a single plat-
form to give a push to solar energy, was jointly launched by the Prime Minister Narendra
Modi and the then French President Francois Hollande in Paris on November 30, 2015.
International Solar Alliance (ISA) was conceived as a coalition of solar resource-rich coun-
tries to address their special energy needs and it will provide a platform to collaborate on
addressing the identified gaps through a common, agreed approach. It will not duplicate or
replicate the efforts that others (such as the International Renewable Energy Agency (IRENA),
Renewable Energy and Energy Efficiency Partnership (REEEP), International Energy Agency
(IEA), Renewable Energy Policy Network for the 21 st Century (REN21), United Nations
bodies, bilateral organizations, etc.) are currently engaged in, but will establish networks and
develop synergies with them and supplement their efforts in a sustainable and focused manner.
The vision and mission of the International Solar Alliance is to give a devoted stage for the
participation of sun-based asset-rich nations, where the worldwide group, including recipro-
983

cal and multilateral associations, corporates, industry, and different partners, can influence
a positive commitment to help to accomplish the shared objectives of expanding the utiliza-
tion of sun-powered vitality in addressing the vitality needs of the forthcoming ISA partner
nations in a sheltered, advantageous, reasonable, evenhanded and practical way.
ISA is intended to be a multicountry partnership organization with membership from
solar resource-rich countries between the two tropics. ISA’s proposed governance structure
would consist of an Assembly and a Secretariat.
3.6 Public–private sector co-operation

Public–private is a form of co-operation between public authorities and the private sector
that aims to modernize the delivery of energy services. This can be a long-term or short-term
contractual relationship between a public entity and a private organization where risks are
shared and there is increased financing of energy efficiency (EE).
Solarization of CIAL
The use of solar energy at airports has developed gradually. Airports experimented with
installations that provided a few hundred kilowatts of peak power at the beginning of this
century. Nowadays, two, five or ten megawatt installations are not uncommon and the eco-
nomics are much improved as grid parity is approached.
Cochin International Airport (CIAL), serving the city of Kochi in the Indian state of
Kerala, is the busiest and largest airport in the state and the fourth busiest in the country. The
airport serves more than five million people annually.
The CIAL (Cochin International Airport Limited) set up a 100 KW PV predicated solar
power plant on the rooftop of the advent block as a pilot project during March 2013. It
required 400 panels, each with a capacity of 250 Wp. The installation was designed and
executed by Kolkata based Vikram Solar Power. This facility engendered 400 units of power
annually, the absence of battery backup reducing the capital cost drastically.
The next logical step was to go for full generation of the power required for the entire
operation of CIAL, integrating up to around 48,000 kilowatt hours per day from a PV predi-
cated solar system in its own backyard. To make the airport grid neutral, the capacity of the
system was to be about 12 MWp, with a capacity to generate 50,000 kilowatt hours. This was
done with permission from KSEB (Kerala State Electricity Board) to bank the electricity.
Bosch Limited was endowed with this esteemed activity through a transparent tender proc-
ess. They did it with élan and with the immaculate specialized flawlessness that Germans
are acclaimed for. The total project cost about `62 Cr. at about `5.17 Cr/MW, which is substan-
tially less than the benchmark set by the regulator. The project payback period is under six years.
M/s Bosch handles the system maintenance as well, on a contract, at a cost of `50 Lac per annum.
3.7 Stake holder engagement in EE initiatives in India

As the major energy consumers, the corporates no doubt have many significant contributions
toward the sustainable development of delivering energy efficiency. The CSR (Corporate Social
Responsibility) can be best implemented toward this. Data analyzed by the Ministry of Corpo-
rate Affairs for the CSR expenditure of all Indian companies in 2014–2015 showed that 14%
(Rs. 1213 Cr) of total CSR spending in India was made on activities focusing on conserving
the environment. It was the third highest expenditure on a social impact issue, after education
(32%) and health (26%), and was greater than the amount spent on rural development (12%).
3.8 Governmental co-ordination

The government’s basic leadership on energy at the central level is appropriated between
the Ministry of Petroleum and Natural Gas, the Ministry of Coal, the Ministry of Non-
Conventional Energy Sources, the Ministry of Environment and Forests, the Ministry of
Atomic Energy, and the Ministry of Power. Inside the Ministry of Power, the Central Elec-
984

tricity Authority (CEA), the specialized wing, works intimately with singular state power
sheets (SEBs) and utilities in the controlling, transmission, and dissemination of power. At
the state level, there exists different divisions, offices and specialists chipping away at differ-
ent sub-segments of energy. Amusingly, the Ministry of Panchayati Raj, working at the town
level, does not have any part to play in rural electrification. The obligation regarding thor-
ough rural electrification (including the quality and collection of power) is scattered between
various Ministries in an awkward manner.
Being a concurrent subject under the Constitution, the states share powers with the center
only for power, not for the other energy sources. However, they do share powers on environ-
mental regulation. This distribution of powers makes co-ordination between the different
energy sources, over the country and between states, difficult.
Environmental clearance is required for all types of power projects, including nuclear,
hydro and thermal, under the Environmental Protection Act 1986 (EIA notification in 1994)
from MoEF, GoI.
3.9 Evaluation
For effective policy-making, compliance evaluation is critical. The importance of evaluation
in policy-making has been well established through the examples of Denmark and Sweden.
Denmark’s energy efficiency program evaluation has made a concerted effort in developing
policy and long-term strategy. Compliance evaluation can assist the Indian policy makers
with identifying potential issues in the execution of the Energy Conservation Building Code
(ECBC) and help them to make necessary changes. Compliance evaluation will likewise enable
India to accomplish its proposed energy savings and emission reductions through the ECBC.
The Indian Energy Security Scenarios (IESS)

The Indian Energy Security Scenarios (IESS), 2047, developed by NITI Aayog (erstwhile
planning commission) is a tool to assess the government’s role in energy use.
This tool is used for tending to requirements, and to give ‘all energy demand’ and ‘all
energy domestic supply’ points of view under different scenarios/suspicions, on a 5 yearly
premise up to 2047 (100th year of India’s independence), with 2012 as the base year. It is used
to teach the nation about its energy status, and enable them to create pathways that India
may embrace for improving its residential energy reliance.
The tool has been used with the assistance of a wide pool of information accomplices from
the government, industry, think tanks, non-governmental associations, international research
organizations and the scholarly community. Broad stake holder consultations were held with
industry, academia, the government, the overall population and different specialists (organi-
zations or people) in various fields of energy to vet the legitimacy of particular directions and
suppositions taken. A few between ecclesiastical counsels inside the government were likewise
directed to validate the information for particular expansive effort workshops have been led
to advance the utilization of this instrument and include more individuals in the activity for
agreement fabricating and making mindfulness about energy arrangements.
3.10 Energy efficiency targets of India

The first cycle of the government’s Perform, Achieve and Trade (PAT) energy efficiency scheme,
which ran from 2012–2015, contributed to an emissions reduction of 31 million tons of CO2,
energy savings of 8.67 million tons of oil equivalent, and avoided capacity addition of about
5.6 gigawatts. The scheme also resulted in monetary savings of Rs. 37685 Cr into energy effi-
ciency technologies by the participating industrial units. The PAT scheme uses a market-based
mechanism to enhance energy efficiency. Large industrial consumers of energy (called desig-
nated consumers or DCs) are given energy efficiency targets. Those who exceed their targets are
awarded energy efficiency savings certificates, which they can sell to those who fail to meet theirs.
At the end of the first cycle, 3.8 million certificates were issued and one ton of oil equivalent
certificates were issued, one ton of oil equivalent being equal to EScert.
985

4 ANALYSIS AND CONCLUSION
This paper has two sections, with the first one being about detailing the concept of energy
efficiency governance in a global aspect. Energy efficiency governance is a new concept that
has evolved linking energy efficiency with governance, which will mainly focus on sustainabil-
ity. For any country to be energy efficient it should implement policies and programs under
the umbrella of the government through governance. India also undertook many initiatives in
energy efficiency indirectly, under the head of energy efficiency governance. This paper intends
to give the energy efficiency framework of the Indian energy efficiency governance. India exhib-
its good practices in public–private participation, international assistance, government co-ordi-
nation, stakeholder engagement, etc., but lacks efficiency in managing the energy sources. India
adopts transparent and accountable systems. The overall institutional governance is weak in
energy efficiency. Separate local level energy efficiency policies have to be created to allow the
actions to begin from the bottom level of governance. There is no constitutional support for
beginning the energy planning at the local level, which will impart energy efficiency. The energy
resourcing and the evaluation of energy efficiency is found to be inefficient in the Indian con-
text. In order to qualitatively analyze the efficiency, not only at the building level but at the area
level also needed so as to confirm to the energy demand and its resource requirements.
ACKNOWLEDGMENT
The authors gratefully acknowledge the contributions of faculty, Government Engineering

College, Thrissur, friends and family for successful completion of the study.
REFERENCES
ACEEE (American Council for an Energy Efficient Economy). (2010). America. State Energy Effi-
ciency Scorecard for 2010, October.
BEE (Bureau of Energy Efficiency). (2010). National mission for enhanced energy efficiency—mission
document: Implementation framework, Ministry of Power, Government of India. agency, I. e. (2015).
India energy outlook. France: Directorate of global energy economics.
FACT SHEET: The United States and India—Moving Forward Together on Climate Change, Clean
Energy, Energy Security, and the Environment (2012, September 22). Retrieved January 12, 5, from
https://in.usembassy.gov/fact-sheet-united-states-india-moving-forward-together-climate-change-
clean-energy-energy-security-environment/.
International Energy Agency. (2010). Hand book of energy efficiency governance. Australia.
International Solar Alliance: Indias brainchild to become a legal entity on Dec 6. (2017, November 14).
Retrieved March 04, 2018, from https://economictimes.indiatimes.com/news/environment/develop-
mental-issues/international-solar-alliance-indias-brainchild-to-become-a-legal-entity-on-dec-6/arti-
cleshow/61647360.cms?utm_source=contetofinterest&utm_medium=text&utm_campaign=cppst.
ISA mission. (2016, August 2). Retrieved January 5, 2018, from http://isolaralliance.org/.
Ministry of New and Renewable Energy (GOI), Ministry of Power (GOI), USAID. (2016). Partnership
to Advance Clean Energy-Deployment (PACE-D) Technical Assistance Program.
Mohan, B. & George, F.P. (2016). Airport solarization: CIAL steals the thunder.
Niti Aayog to rank states on energy efficiency. (2017, February 22). Retrieved March 04, 2018, from
http://economictimes.indiatimes.com/industry/energy/power/niti-aayog-to-rank-states-on-energy-
efficiency/articleshow/57301449.cms.
Rao, S.L. (2012). Coordination in energy sector and its regulation in India. Institute for social and eco-
nomic change. 113–120.
Reddy, B.S. (2014). Measuring and evaluating energy security and sustainability: A case study of India.
Statistics. (2017, December 5). Retrieved January 5, 2018, from https://www.iea.org/publications/.
Together on Climate Change, Clean Energy, Energy Security, and the Environment. (2012, September
22). Retrieved January 12, 5, from https://in.usembassy.gov/fact-sheet-united-states-india-moving-
forward-together-climate-change-clean-energy-energy-security-environment/.
Yu, S., Evans, M. & Delgado, A. (2014). Building energy efficiency in India: Compliance evaluation of
energy conservation building code. U.S Department of Energy.
986

Environmental management of a blue-green network: A case

of Valapad in Thrissur, Kerala
Nahlah Basheer
Masters of Urban Planning, Government Engineering College, Thrissur, Kerala, India
C.A. Bindu
ABSTRACT: Natural resources and the extent of their management are important for any
region, particularly to meet the demand for resources in these times of change. There should
be a ground for effective planning and management of these resources. They have environ-
mental, ecological, socio-cultural and economic roles to play among many others. The need
for an integrated development plan for Valapad, a coastal gramapanchayat of Thrissur dis-
trict in Kerala, India highlighted a requirement for a detailed on-site study of the environ-
mental sector of the area. The focus was on developing effective water management and
green infrastructure simultaneously. The changes that have taken place in the land use due
to the changes in land utilization, levels of encroachment, low public awareness, destruction
of flora and fauna, for example have contributed toward diverse effects that are irreversible.
This paper focuses on the amount of environmentally sensitive areas present in the region,
the existing scenario and the strategies that can be adopted for the management of the same.
1 INTRODUCTION
Long term sustainability of any urban or semi urban area is often related to that of economic
growth. But in reality it is crucially dependent on the social, economic and environmental
dimensions of the environmental sector. A multidimensional process like planning and devel-
opment requires an in-depth probe into matters of environment and protection. The case of
Valapad, a coastal gramapanchayat in the Thrissur district of the south Indian state of Kerala,
studied for the formulation of an integrated development plan, is explored in the paper.
2 REGIONAL SETTING
Valapad gramapanchayat lies in the Manappuram region of the west coast in Thrissur,
which has a separate island-like formation. This seaboard tract extends from Chettuva in the
north, to Azhikode (Munambam) in the south. The total length of the belt is approximately
56 km and the total width varies from four to eight km. The water bodies around the belt
are Karuvannur River, Canoli Canal and the Arabian Sea. The Karuvannur River flows
encircling this formation in the northern side, joined by the Canoli Canal from almost the
mid portion, flowing southwards. When it reaches the Munambam region, it is flushed by the
adjoining waters of the Periyar River. On the western side of the land is the Arabian Sea. It
is this island-like encapsulated land that we call the Manappuram region. Valapad is situated
toward the middle of the Manappuram region. The approximate width observed is 5 km.
Valapad followed a ridge and valley terrain in topography where valley areas were drained
throughout by natural streams (known locally as thödu). Some of these were fit for travel.
Other valley areas were wetlands and cultivable paddy fields. The depressions were filled with
987

natural ponds. Today, almost all of the valley areas of yesteryears, such as the wetlands and,
paddy fields, have undergone land filling and most of the natural drains have become unfit
for regular water passage. This has brought about a considerable change in the topographi-
cal pattern of the region. The area now consists of almost flat land. There is no considerable
slope observed. The type of soil is sandy, which is acidic in character. So, during the summer,
frequent irrigation is a must in the region.
Valapad has a well-established road network comprising of a National Highway, Major
District Road and Village roads. It has a transitional character in its location as it is located
in between several urban centers. NH 66 lies on the eastern side of the panchayat and hence,
the road network forms a grid iron pattern here. This gives rise to easy inter as well as intra
connectivity within and outside the region.
3 NATURAL RESOURCES IN VALAPAD
3.1 Pond
There are numerous ponds in the gramapanchayat area and it is therefore a complete natu-
ral ecosystem. In earlier times, almost all households had one or more ponds, of which one
source was for bathing and the other for drinking. Most of the ponds of Valapad are mapped
in Figure 2, with Muriyathoodukulam, Kothakulam, Ambalakkulam and, Thirunellikkulam
being the important ponds in the area. Although there are numerous ponds here, most of
these are not in good condition.
Figure 1 details a few ponds by their nature of ownership (for example private, public and
temple ponds), their usage, features and pollution rates. It should be noted that the com-
monly observed threats to ponds here are pollution, land use changes, climate change, inef-
ficient water management techniques, improper managements of pond water, intensive use
for irrigation, fish overstocking, and degraded buffers.
3.2 Wetlands and paddy fields

The Nancha and Vayal areas of the panchayat have been showing a dwindling trend for a long
time. Agriculture is no longer a source of income here. According to the present statistics,
only seven hectares of land in the panchayat is under paddy category. This area is not used for
paddy cultivation now. The Kuttippaadam in Ward 15 of the panchayat is one such example.
3.3 Sacred groves

Sacred groves are a very ancient and widespread phenomenon, and there are numerous sacred
groves in Valapad panchayat, although it should be noted that there are only three to four
Figure 1. Ponds: Usage, features and pollution rates.
988

Figure 2. Location: Existing ponds and wetlands/paddy fields in Valapad gramapanchayat.
Figure 3. Used and the majority of unused ponds in Valapad.
Figure 4. Some sacred groves of Valapad gramapanchayat (a) Thekkiniyedath Naagakkaavu
(b) Arayamparambil Kaavu (c) Paarekkaatt Kaavu (d) Cheeramkaattil Kaavu.
well maintained ones. All the others are cleared of the lush vegetation and reduced to a single
platform for worship. The few notable ones are shown in Figure 4. Of these, the Adipparam-
bil Kaavu (Nagayakhi—Sarpakkaavu) of the Arayamparambil family is the richest with more
than 100 varieties of medicinal plants and other flora in the thirty cent premises. According to
the studies conducted by Jincy, T.S. and Subin, M.P. (2015), the area of this sacred grove has
been reduced to thirty cents from eighty-four cents in 1998, following land partition.
3.4 Streams
The 3,880 m long Pannatthödu, the 5,320 m long Paalamthödu, the 7640 m long beach thödu,
the 1,200 m long Netkot thödu, and the 9620 m long Arappathodu flowing through wards are
the most important streams in Valapad. However, there are other multiple feeder and connec-
tor streams to each of these main streams. Streams are numerous but with respect to the flow
of water and connectivity, the case exhibits a rather degrading scenario. Neglect and unre-
stricted waste dumping has resulted in blockage. Restricted water flow during the monsoons
lead to flooding of streams and as a result, the people resort to unscientific water clearances.
Known as the breaking of Arappa (estuary), the rainwater collected and flooding the lands,
is allowed to flow directly into the sea, thereby, giving no chance for groundwater retention.
3.5 Mangroves
Thrissur district consists of very low numbers of mangrove in the state. Presently, mangroves
are confined to the backwaters of Chettuwai, Azhikkodu, Kodungallur and few patches in
Venkidang and Pavaratty Panchayats. Valapad is abundant with an ecosystem favorable for
the growth of mangroves. Mangroves are present along the Kothakulam Arappa area, bor-
dering the Kothakulam beach ward, located approximately 500 m away from the sea. This is
989

Figure 5. (a) Sacred groves and mangroves (b) stream network.
Figure 6. Existing condition of streams: Waste accumulation hindering flow.
one of the most picturesque spots in the panchayat. However, there are no buffers against the
mangroves and sooner or later, it may suffer from encroachment as elsewhere.
3.6 Surface water conditions

The surface water sources in Valapad gramapanchayat are the water from the Karuvannur
River, Periyar River, and Canoli Canal. The Karuvannur River is the fourth largest river in
Thrissur district, formed by the confluence of the Manali and Karumali rivers.
The instruments of natural surface water collection are the natural ponds, Arappa, and
the various natural drains or the thödu. The Valapad manappuram region comes as a micro
watershed of Karuvannur puzha. Hence, it should be noted that the only sources of water to
the water bodies is from the surface run off and rainfall.
The Arappathodu once filled, leads to stagnation issues and many wards suffer. Hence, con-
nectivity issues arise. As already mentioned (in section 3.4) residents resort to breaking the
waters. The collected water is immediately drawn to the sea. There would be no collected water
for groundwater recharge. This has led to the lowering of the freshwater table. Water recharg-
ing is the only possible way for the regeneration of water to the waterbodies in the panchayat.
In Valapad, there are no rainwater collection pits, no rooftop rainwater harvesting systems:
neither at household levels nor at panchayat levels. From the primary survey conducted, it was
found that only 2% of the people have set a rainwater harvesting system in Valapad. Apart from
the existing saltwater intrusion, this may lead to acute water shortage in the near future.
3.7 Biodiversity
Being a coastal panchayat, Valapad has the potential to serve a large amount of aquatic and
terrestrial biodiversity. The changes in the land use and encroachment over the protected
areas can be attributed to the depleting wilderness and ecologically sensitive areas. These
lead to endangered plant and animal life. There is also reduction in the amount of medicinal
plants that were once easily and largely available in the panchayat.
990

Figure 7. Surface water and shortage.
Figure 8. Water problems: Groundwater and aquifer details.
4 LINKAGES WITH OTHER DEVELOPMENT SECTORS
In Valapad gramapanchayat, the environment related issues of water shortage, pollution, and
waste management are also linked with the shortage of physical infrastructure. The problems
of waste management are acute and this has in turn led the people to resort to means of
burning plastic wastes and also dumping them into the water bodies. This pollutes the waters
and also disrupts the connectivity of flowing water in the streams. Pollution in the form of
air and noise along the NH at peak traffic hours is one of the negative effects of traffic and
transportation on the environment sector.
Beach and beachfront ambience is the main attraction of Valapad for tourism. Incorpo-
rating the principles of the environment and ecology to suit nature tourism, responsible eco-
tourism projects can take up momentum in the area. Integrated tourism planning is essential
for this and such initiatives can help in the sustainable development of the area.
5 STRATEGIES FOR REVIVAL
The strategies pertaining to the protection, conservation and maintenance of the natural
resources that shall be adopted in Valapad, includes those related to ponds, streams, sacred
groves, mangroves, flora and fauna, groundwater recharge, waste management and so on, for
a better living environment. In the case of Valapad, bringing reasonable participation of the
stakeholders, through early and effective consultations by all partners is crucial to frame a
partnership. The actions required for this effective management shall be further exercised on
three levels, as shown below in Figure 9.
Ponds: Though Valapad is known to have thousands of ponds within the gramapanchayat,
the area lacks an action plan to maintain these assets. Table 2 below gives relevant strategies
that can be adopted for effective maintenance of ponds and the lifeline of water retention in
the panchayat.
Streams: Table 3 below gives relevant strategies that can be adopted for effective mainte-
nance of streams here. For any action, the existing conditions of water channels and their
connectivity need to be studied and mapped.
991

Figure 9. Levels of management to be exercised.
Table 1. Focus areas at each level of management plan.
At level Focus areas shall be

Levels
1 Physical changes and improvements Infrastructure, planting, and asset management.

2 Policy and planning Changes in land use, infrastructure and movement.
3 Implementation agency plan for Identification of agencies for sectoral mix in order to
sectoral intermix consider the strategies for an execution plan.
Table 2. Strategies for maintaining ponds.
Levels Strategies
1 Regular cleaning (especially of the ponds used for the holy dip of Aaraattpooram, like
Kothakulam, and Sethukulam).
2 Resource mapping of existing unfilled ponds. Curb on unscientific landfilling of
waterbody. Introduce functional interfaces along public ponds in a regulated manner.
3 Device a maintenance partnership plan through efforts from beneficiaries (private owners)
and panchayat at a reasonable share.
Table 3. Strategies for maintaining streams.
Levels Strategies
1 Regular cleaning (especially of the feeder streams to the Arappathodu).

2 Network-mapping of existing streams (feeders and connectors included).
Redefining of edge character with buffers to curb waste disposal into the waters (resulting in
blockage).
Introducing functional interfaces promoting public open spaces and regulated tourism
activity along Arappathodu parts (for instance, suitable in Kothakulam).
Integrating streams to be a known and usable part of the panchayat.
3 Devising a management plan incorporating MGNREGA (Mahatma Gandhi National
Rural Employment Guarantee Act) workers to solve connectivity issues. Private
partnership may not be viable.
Sacred Groves: The present conditions throw light on the fact that sacred groves of Valapad
have attained religious recognition but no environmental recognition. As such, due to strong
religious beliefs surrounding the same, private owners put efforts into their conservation.
Only as it moves forward to attain the level of an environmentally sensitive spot, will the asset
become a social responsibility.
Mangroves: Mangroves are one group of natural elements which aid in many ways against
environmental and natural hazards. Table 5 gives relevant strategies that can be adopted for
their conservation in Valapad gramapachayat.
Medicinal Plants: Table 6 below lists relevant strategies that can be adopted for the revival
of medicinal plants that once adorned the panchayat area.
Water Shortage: The integration of projects like Mazhapolima with a cost managed like
other sample areas of Guruvayur Municipality, Engandiyoor Panchayat and others. A
plan such that 25–30% of the project cost (for an individual household) be provided by the
beneficiary contribution shall be proposed. The beneficiary list once submitted to the
992

Table 4. Strategies for conserving sacred groves.
Levels Strategies
1 Promote afforestation of medicinal plants and trees inside the groves for better
environmental stability.
Manage a data bank of existing well managed sacred groves in the gramapanchayat by
laying them out into categories of size, density of vegetation and expanse.
2 General awareness about the social and environmental need for conservation (apart from
the religious concepts) and existing assistance schemes.
Spell out protection and conservation projects in the panchayat by providing funding
from the local Panchayat funding. There ought to be protection using buffers (natural/
artificial) according to the data gathered.
3 Devise a conservation plan for the principle of partnership with efforts from beneficiaries
(private owners), local communities around the nearest vicinity and local authority
(panchayat) at a reasonable share.
Table 5. Strategies for conserving mangroves.
Levels Strategies
1 Boost the existing and initiate planting of new saplings in the Kothakulam Arappa area, at
the same time, protect the already planted sapling areas in Palappetty.
2 Identify feasible zones for the planting of mangroves
3 Devise a maintenance partnership plan through efforts from beneficiaries (private
owners) and panchayat at a reasonable share
Table 6. Strategies for reviving medicinal plants.
Levels Strategies
1 Through clearing large sites of neglect in each locality, there shall be enough space for
planting.
2 Identify suitable areas of potential for a botanical medicinal plant garden.
3 MGNREGA workers and youth clubs can work together to site setting. Devise a manage-
ment plan involving the women folk of Kudumbashree, environment clubs of schools,
Government Ayurveda Hospital Valapad and the many practitioners of naturopathic
therapy.
gramapanchayat; a technical team of the project can provide the necessary support for the
installation of the open well recharge units.
For initiating the project, the panchayat area can be broken down into clusters and these
clusters can be selected based on land use or population density. For example, if on the basis
of land use, clusters can be as in Figure 10 below, namely cluster 1 (residential), cluster 2 (resi-
dential, commercial/industrial) and cluster 3 (residential, commercial/industrial, public/semi-
public use) etc. It shall be seen that in cluster 1 itself, there are three different zones in varying
scales. In stage 1, a small set of houses can be selected for initiating the project. In the next
three months, the stage 2 zone can be integrated with the already existing trial area. Finally, the
final set of the stage 3 zone can be added within the next five months to complete a complete
zone. The need to expand the zones one by one is very important in this project because from
previous successful projects we know that, the success rates of groundwater recharging can
only be achieved from collective participation and involvement from a wide area. Within a
span of two years, the whole panchayat area shall be successfully integrated under the project.
Waste Management: To be commenced at the household levels. With most households with
ample land parcels being able to manage their own food wastes and replacing plastic with
reusable/biodegradable materials wherever possible can be linked to better results. Also, scope for
the use of modern technologies such as those in bailing machines, will exist in the near future.
993

Figure 10. Proposed clusters for rainwater harvesting.
6 CONCLUSION
The cases of land and water pollution are acute in Valapad, due to the improper and unsci-
entific waste disposal techniques adopted. However, there are instances of certain public
ponds being cleaned by schedule under the MGNREGA works. But at the household levels,
most ponds are used for waste dumping and left for filling. In terms of disaster susceptibility,
the coastal areas along the Kothakulam and Nattika Arappa (estuary) portions are prone to
storm surges and accretion. In a time of reduced rainfall and drought, this is a very serious
problem thet requires higher regard and concern. The existence of a strong blue-green net-
work is a backbone for the development of any area. The management and protection of the
same through effective planning strategies and policies will help strengthen not only the life
of these networks, but also of the people of Valapad, especially because of the multiple ben-
efits these elements shall serve. This is indeed a vision toward an ultimate living environment
for the present as well as generations to come.
REFERENCES
Biodiversity Register. (2010). Flora and fauna. -Valapad: Valapad gramapanchayat.

CWRDM. (2013). Impact assessment of Mazhapolima Project on groundwater regime in Thrissur District.
Interview by Numerous interviewers. (2016). Primary survey on Integrated development plan for
Valapad.
Mohandas, M. & Lekshmi, T. (2012). Kerala mangroves—pastures of estuaries—their present status
and challenges. IJSR.
Subin, M.P. & Jincy, T.S. (2015). A detailed survey on sacred groves of different extends in the Coastal
Belt of Thrissur District, Kerala, India. Global Journal for Research Analyses 163.
Sudheer, M. (2016). Interview by Nahlah Basheer. Water issues in Valapad gramapanchayat.
Valapad K.B. (2016). Interview by N. Basheer-Siji Mohan. Ongoing plan for natural resource mainte-
nance at Valapad.
994

Urban metabolism and food security: Emergy as a metrics link

connecting the food security with the urban spatial aspects for
enhanced livability and sustainability
Bijey Narayan & J. Jayakumar

Department of Architecture, College of Architecture Trivandrum, Trivandrum, India
ABSTRACT: Urban food insecurity is a major challenge associated with the phenomenon
of global urbanization. Several such multi-scale socioecological challenges have necessitated
the re-emergence of the concept of urban metabolism, which essentially deals with the flow
of energy and materials into and out of the city. In cities, the production and processing of
food, though being away from the consumers, its consumption and disposal are still with
them, which shows there is an obvious rift in the urban metabolism in terms of the food flow.
The authors, with the help of the emergy model establish a strong link between urban food
systems and urban land use pattern, highlighting the importance of sustainable urban plan-
ning that is also sustainable from social, economic and ecological perspectives. This model
envisages a city that produces its own food and nutrition as well as distributes, consumes and
disposes of it by virtue of the efficient systems generated by the urban resources.
Keywords: urban metabolism, emergy model, emergy analysis, urban agriculture, closed-
loop food system, urban food security, urban land resources, urban land use plan, urban
planning
1 INTRODUCTION
According to UN reports on world urbanization prospects, around 66% of the world’s popu-
lation will live in cities by the year 2050. The majority of this urban growth will take place
in low and middle income countries. The projected urbanization shows an alarming rate
and scale that can raise slum populations to 2 billion people (United Nations Report, 2014).
Urban planning, design and governance are becoming centrally critical for human survival.
The statistical evidence is in general pointing to the fact that humans as socialized animals
are inhabiting smaller land parcels where the whole of the existing urban areas is only around
1–3% of the land area of the earth. The non-convertible wastes generated here are also solely
by humans. All human made systems inadvertently over-depend or exploit nature, consider-
ing it as the inexhaustible vessel to meet their material needs.
The urban data also mentions another unpleasant fact that many of the city-dwellers, par-
ticularly those living in slums, still suffer from malnutrition (Amerinjan, 2017). Urban dwell-
ers, especially urban poor are less privileged in terms of the accessibility to healthy nutritious
food than rural people. This can be attributed to many factors, including the fast pace of urban
life and to the skewed socioeconomic bias of the urban land use pattern. It is considered that
the economy of a city depends mostly on the income generated by the business, commercial
and industrial activities taking place in its built-up land. Hence incorporating or retaining
green areas in the urbanized areas has less priority and consideration. As production of food
is possible primarily with agricultural land being made available along with labor, skill and
infrastructure facilities, urban areas by default become consumers of imported rural food
and highly processed agricultural products. In general, the high energy transportation cost
of food products from outside, which in many cases is another continent, drastically reduces
995

the required accessibility. Hence much needs to change in today’s urban spatial structure and
organization in order to keep cities fed with healthy and nutritious food.
In countries like India, where the pattern of development has always been haphazard and
equally unpredictable, the process of restructuring urban spatial organization will be quite
complex. The majority of the medium cities of India lack a land use plan even today (Niti
AAayog, 2017). Considering the unprecedented trends in urbanization, a mechanism or
model looks inevitable to relate the needs of the urban dwellers to their consumption pattern
(material flow into and out of the cities) that can be converted to a value which in turn forms
the key parameter for an urban resource allocation strategy, especially urban land resource
planning. The authors are suggesting incorporating a less data intensive tool to assess, esti-
mate and balance the inputs and output in the urban system.
Food security being the focus of this paper, the exploration here is into the flow of food
material in the urban areas and its relationship to urban land resource allocation strategies.
2 FOOD SECURITY
Definition of food security has changed over the years. It was defined in the 1974 World Food
Summit as the ‘availability at all times of adequate world food supplies of basic foodstuffs
to sustain a steady expansion of food consumption and to offset fluctuations in production
and prices’ (FAO, 2003). In 1983, Food and Agricultural Organization of the United Nations
(FAO) defined it as ‘ensuring that all people at all times have both physical and economic
access to the basic food that they need’ and thus expanded the concept to include access by
vulnerable communities or people to available supplies, implying that the demand and sup-
ply side of the food security equation should be balanced. The 1994 United Nations Devel-
opment Program (UNDP) Human Development Report added a number of component
aspects to human security of which food security was only one. This concept, that related
food to the human rights, helped start discussions about food security. The 1996 World Food
Summit adopted a still more complex definition: ‘Food security, at the individual, household,
national, regional and global levels [is achieved] when all people, at all times, have physical
and economic access to sufficient, safe and nutritious food to meet their dietary needs and
food preferences for an active and healthy life’ (FAO, 2003).
The State of Food Insecurity refined the definition in 2001 as: ‘Food security [is] a situ-
ation that exists when all people, at all times, have physical, social and economic access to
sufficient, safe and nutritious food that meets their dietary needs and food preferences for an
active and healthy life’ (FAO, 2003).
Hence the accepted definition of food security can be expressed as: food security exists
when all people, at all times, have physical, social and economic access to sufficient, safe and
nutritious food that meets their dietary needs and food preferences for an active and healthy
life. Household food security is the application of this concept to the family level, with indi-
viduals within households as the focus of concern, and ‘Food insecurity exists when people
do not have adequate physical, social or economic access to food as defined above’ (FAO,
2003). In the state of Kerala, food insecurity has become a huge challenge because of the
toxic content and questionable safety of the food imported from the neighboring states. Of
late, Kerala has witnessed a change in the vegetable production in its urban areas intended for
local consumption rather than relying on the food imports from distant places.
3 URBAN INDIA
According to the 2011 census, 377 millions of Indian people live in urban areas compared to
286 millions in 2001. For the first time since independence, the absolute increase in popula-
tion is more in urban areas that in rural areas (Census India, 2011). A study of Indian cities
by McKinsey Global Institute, reveals that the current performance of Indian cities is poor
across key indicators of quality of life, such as water supply quantity, public transportation,
996

Figure 1.
Source: UN DESA Rome, 2010.
Figure 2.
*United Nations Department of Economic and Social Affairs.
Source: *UN DESA Rome, 2010.
parks and open spaces. On current trends the quality of urban services is expected to fall
further by 2030 (McKinsey, 2010) with reasons obvious even today.
The mentioned scale and complexity of urbanization in India demands a comprehensive
strategy required in addressing the urban challenges, especially that of food security. Inten-
sive data preparation is necessary to devise strategies for urban resource allocations and the
use of accurate metrics is required to ensure the effective execution of urban strategies.
4 LAND USE PLANNING IN INDIA
Town or urban planning in India is still in its infancy. While an urban master plan governs
the growth of the urban areas, a Town Planning Scheme or TP Scheme is also widely adopted
in various states. Land use planning, the most important feature of all development plans
dealing with spatial planning, is very critical for all development purposes since it involves
assigning the particular activity to a given parcel of land based on the concept of optimal
land rent where the very definition of optimal use is limited to economics. The authors are of
the opinion that there has to be another dimension to it, by adding the component of solar
energy falling on the urban land surface for the production of food from within, which is
important for sustainable development.
Zoning and land use planning help to ensure dedicated land for segregated and planned
activities that have positive impacts on the economy and to avoid conflict in activities that
affect the quality of life in those areas. Unfortunately many of the fast urbanizing, middle
997

sized towns do not have a comprehensive master plan with a well-developed land use plan
(Niti AAayog, 2017). As there is not much credible work done in the urban land use plans
or the development of a master plan in the various cities and towns of India (Niti AAayog,
2017), it is imperative that there has to be an alternative system of metrics to link and meas-
ure the needs, flow and use of urban resources and materials.
5 URBAN METABOLISM
A city can be better understood if compared with a complex organism with various meta-
bolic processes. Howard T. Odum, an American ecologist, proposed the conceptual model
of metabolism (Odum, 1996) where materials from within the city or from outside getting
transformed by a series of urban activities, and finally get converted to waste and then
released into the environment. Urban metabolism is composed of built-up land, farmland,
and unused land (Odum, 1996).
The inner environment of cities by itself cannot support all the metabolic activities; mate-
rials, and energy from outside is also needed for it. Additional mechanisms are required to
expel the wastes into the environment. Hence urban metabolism is understood and expressed
in terms of production, consumption, and processing of urban internal resources completed
with waste disposal, as well as the flow of material and energy between the internal and exter-
nal environment (Odum, 1996).
6 URBAN METABOLISM, URBAN FOOD SYSTEM AND FOOD SECURITY
Urban Food security depends on the design and resilience of the urban food system, which
in turn depends on the urban metabolism. Metabolic rifts are caused due to the linear, inter-
rupted food loop systems, which happens due to the exclusion of food from the urban policy
and planning agenda for a long time.
The urban food system has a dynamic structure that exists between land, population, food
distribution and production processes, resources, technology, economy and employment
(Armendáriz et al, 2016).
A resilient urban food system addresses all four levels of the food system: food production,
processing, distribution and consumption, (Amerinjan, 2017) which implies a closed-loop food
system. In the cities, policies can be established to incentivize the local production of healthier
food options and limit unhealthy food imports (Amerinjan, 2017). The Edo period in Japan
could be a good model for this. Recycling and living with ‘just enough’ were made part of pub-
lic policy when Edo was facing an imminent environmental crisis (Brown, 2010). Almost every
material in Edo was made to recycle and none went to waste. The need to recycle was reflected
in technologies and practices and the way things were made in the first place. This could argu-
ably be the predecessor of the circular economy that we are trying to engineer today. Similarly,
we need a new paradigm within the urban planning that will help ensure food security and
thereby sustainability in the cities. Metric-based urban planning is required where the urban
metabolic density can be evaluated and allocation of land resources can be done accordingly.
7 ‘EMERGY’ AS METABOLIC METRICS
Emergy is defined as the total amount of available energy (or exergy) of one kind that is used up
directly or indirectly in a process to deliver an output, product, flow, or service (Odum, 1996).
Emergy analysis is a method to measure the value or quantity of material energy that is used
to transform the different (2 or uniform measurement standard), by the use of specific conver-
sion factors and by combining socio economic with eco-environmental systems, to analyze the
flows and transformations of materials and energy quantitatively (Huang et al, 2015).
The formula is given as
Em = τEx
998

Figure 3. Scenario journal 15.
where Em is the emergy of one material or energy, Ex is the available joules of one material
or energy and τ is the emergy transformity constant of material energy, which is the solar
emjoules it needs to transform (Huang et al, 2015).
According to the emergy model, different forms of energy, materials, human labor and
economic services are all evaluated on a common basis (the environmental support provided
by the biosphere) by converting them into equivalents of only one form of available energy,
the solar kind, expressed as solar equivalent Joule (seJ).
The concept of “available energy” allows the analyst to account for all kinds of resources
used (minerals, water, organic matter), not only energy carriers (Huang et al, 2015).
A typical material flow study can be seen from the figure. Food materials can be filtered
from it and studied separately. Using the emergy database, the solar transformity value can be
found for each material, and that can be multiplied with the available amount of the material
to find its emergy. Various materials and energy are categorized into different groups for ease
and standardization in calculation, such as renewable, non-renewable, industrial and labor.
The solar transformity of some materials are as given below.
1. Sunlight = 1
2. Agricultural production = 1.43 × 105
3. Livestock production = 9.15 × 105
4. Fisheries production = 3.36 × 106
In this way, the emergy-based evaluation of urban land can also be done. By calculating
the correlation coefficient of the increments in various urban lands, such as built-up land,
its impact on the total metabolic density of the city can be found out. If a city’s metabolic
activities depend on the activities in the built-up land as per land use, it can be concluded
that the changes in the activities on the built-up land will affect the urban development more
significantly than the other lands. Considering the environmental load of the built-up land,
planners can insist on more farmlands/agricultural land to be included in the urban master
plan, to match the emergy value of the food materials coming in and required to be disposed
of in the urban area. To address the space shortage in cities, appropriate policies may be
designed and implemented to make sure that there are enough sun exposed roof spaces for
urban agriculture/farming. This will further necessitate the need for scientific disposal and
recycling of wastes that can in turn be used for the production of food items in the urban area
itself. Thus, urban food systems can be made a closed-loop system, fostering sustainability
and resilience and eventually, a food secure urban community.
8 CONCLUSION
This paper has conceptually explored emergy as a metrics parameter in urban resource
allocation to achieve urban food security. Emergy is useful in evaluating and relating the
999

environmental and economic aspects of a triple bottom line development model but has its
limitations in evaluating the social component. However, considering the rapid urbanization,
urban migration, environmental deterioration and increase in GDP in India, a metrics driven
execution of urban strategies definitely has an advantage in the development of cities that
are data deficient.
REFERENCES
Aayog, N. (2017). Documents, Government of India.

Armendáriz, V., Armenia S. & Atzori, A.S. Systemic analysis of food supply and distribution, systems
in city-region systems—An examination of FAO’s policy guidelines towards sustainable agri-food
systems (December 7, 2016).
Brown, A. (2010). Just enough: Lessons in living green from traditional Japan. Tokyo: Kodansha Interna-
tional Ltd. Retrieved from http://sedac.ciesin.columbia.edu/data/collection/gpw-v4.
Census India (2011). Reports. Retrieved from http://censusindia.gov.in/2011-prov-results/paper2/data_
files/india/Rural_Urban_2011.pdf.
Food and Agricultural Organization of the United Nations, Rome (2003). Food Security: Concept and
measurements. Retrieved from http://www.fao.org/docrep/005/y4671e/y4671e06.htm.
Food and Agricultural Organization of the United Nations. (2010). Policy Brief 10. Retrieved from
http://www.fao.org/economic/es-policybriefs/multimedia0/presentation-urban-agriculture-and-food-
security/en/.
Food and Agricultural Organization of the United Nations. Urban agriculture (n.d) Retrieved from
http://www.fao.org/urban-agriculture/en/.
https://esa.un.org/unpd/wup/publications/files/wup2014-highlights.pdf.
Huang, Q., Zheng, X. & Hu, Y. (2015). Analysis of land-use emergy indicators based on urban metabolism:
A case study for Beijing. Sustainability. Retrieved from http://www.mdpi.com/journal/sustainability.
McKinsey Global Institute. (April, 2010). Report on India’s urban awakening: Building inclusive cities,
sustaining economic growth.
Odum, H.T. (1996). Environmental accounting: Emergy and environmental decision making. NY, USA:
John Wiley and Sons Inc.
Van Ameringen, M. (2017). Global alliance for improved nutrition. [Blog]. Retrieved from https://www.
gainhealth.org/knowledge-centre/urban-food-systems-that-promote-good-nutrition/.
Wiskerke J.S.C. (2014). Urban food systems. World urbanization prospects: United Nations report (pp.
1–25).
1000

Smart city initiatives

Smart-cities and smart-villages in the Indian context:

Some behavioral aspects
P. Varghese
Karpagam University, Coimbatore, Tamil Nadu, India
ABSTRACT: Questions have been asked regarding the feasibility of smart-cities; it has
been asked whether the policies could be lopsided when seen alongside normally developing
urban processes, which adopt preconceptions or a uniform distribution of resources, or
next to traditional lifestyles. Some assumptions adopted by Glaeser (2011), as well as the
McKinsey report (2010), regarding urbanization are reviewed, especially within the Indian
context. This needs to be seen within a historical/traditional perspective and does not fully
require the smart-city development policy as a given. It is argued instead that providing urban
amenities in rural areas could also be an alternative or parallel option, for multiple reasons.
1 INTRODUCTION
1.1 The future of urban India

The process of urbanization has been stated to be a real and continuing phenomenon the
world over; in India it has been recorded at about 31% as of 2011, is projected to be 50%
by about 2030, of 590 million, and 70% by 2050 (UNFPA, 2007). As per the last census,
377 million persons (31.2% of the population) live in urban areas (GoI, 2011). It is expected
that by 2031, approximately 590 million persons (40%) will be living in urban areas. In 1951,
only 6.2% of the population were living in urban areas—this change represents approximately
a sixfold increase; meanwhile, towns and urban areas have doubled in number (Ravi et al.,
2016).
Urbanization as a naturally occurring phenomenon cannot be negated or overlooked.
Current urban perspectives are looked at through the arguments of the economic urbanist
Glaeser (2011) in his book The Triumph of the City; he believes that compact cities optimize
real estate prices and reduce congestion, slums and sprawl. A similar argument for increasing
urbanization, the McKinsey report (2010), forms the basis for the development of smart-
cities in India.
It is not being argued here that normal rates of urbanization need to be hindered, which
is not feasible within democratic set ups; it is argued that urbanization has to be undertaken,
especially considering that current urban requirements often lack standard infrastructure or
amenities at minimal levels.
What is being questioned, however, is the development of the so-called smart-cities pro-
gram, which allocates funds to chosen cities as opposed to all cities, which also need similar
funds for much needed infrastructure and developmental works.
1.2 The needs of rural India

The needs of the rural parts of the country are real; India primarily builds on its agricultural
base, where people have been farming for ages. This background is not merely for economic
ends, but there is an existing attachment to the land, to nature, to the environment and to the
idea of continuity. There could be a cultural bias, but it can be argued that it is an ethos that
1003

drives the ordinary man to the land and to nature. Glaeser’s (2011) perspective is certainly
humanistic, despite him being a rational economist, and he is real in his logical arguments. But
it probably does not accept the reality of generations of farmers who have adopted this as a
way of life. It is easy to rationalize that all of existence should be viewed from the perspective
of demand and supply, which would be the base that economists would adopt. This is perceived
only by visible, practical aspects that probably miss the deeper links, which need a sociological
perspective and analysis to also come up with theories of an urban bias by policymakers.
1.3 Non-economic perspectives of the urban processes

Glaeser’s (2011) point of view is logical in stating that the need for development in cities is
separate from rural areas. The McKinsey report (2010) is a pure, rational, economic argu-
ment that projects current global trends, and extends the view that the same holds good for
India. One cannot fault the views from a logical standpoint, which shows realism, but it is not
necessarily the only and definitive perspective.
Instead, alternative socio-cultural possibilities also need to be taken into account, since
it almost assumes that urbanization is the answer to rural poverty; possibly even stemming
from failed agricultural investments, which depend substantially on the monsoon and on
short-term weather predictions. It is being argued instead that providing urban amenities
in rural areas would aid rural folk to continue to remain there, and additionally, help pre-
serve traditional knowledge of the world, of nature and of societies that have continued over
generations. The use of modern technology to document and analyze traditional wisdom,
culture and lifestyles can be prioritized from this viewpoint.
2 URBAN DEVELOPMENT IN INDIA
2.1 Urbanization in India

Currently, India has an annual nominal per capita GDP of US$ 1,408/- (∼Rs. 93,000/-) and
approximately 363 million persons (30% of the population) live in poverty, about 1.77 million
persons are homeless, and 4.9% of the population (aged 15+) are unemployed, while even
those who are employed do not have stable incomes. In the rural areas, where wages are even
lower, possibly by up to half, and more so for women than for men, about 48% of households
survive with little or no socio-economic services (UNFPA, 2007).
Urban growth in India has not developed to the uniform extent that it should have due to
several conflicting factors. The regions of the country have been shaped over time by various
characteristics: the history, topography and climate, as well as traditional and cultural norms.
With regards to the economy, central and state allocation of funds were the main sources
until the recent adoption of the 74th Amendment, when urban areas were expected to build
their own reserves. Strangely, even today there are several bodies within a single area that
have overlapping jurisdictions and work in unco-ordinated ways. This would inevitably lead
to the duplication of administration, unco-ordinated work, a lack of responsibility, inef-
ficiency, and wastage.
A major factor that is lacking is the dearth of qualified officials with a background in areas
such as physical planning, transportation, etc.; bureaucracy and a lack of co-ordination
seems to plague the official administration. Much of this happens at a pace that is prone to
delays, and the adoption of current technological/social media that would help to keep pace
with needs is not looked into; training for such things is possibly lacking. The whole official
machinery and bureaucratic system needs a rethink and an overhaul to be up-to-date.
In such settings, perhaps the world over, it is no surprise that private enterprise takes up the
initiative to create socio-economic change, but with a profit motive; the private sector runs
by eliminating redundancy and wastage of resources and time. The judicial processes are also
long and drawn out, often taking years for a decision.
Governance in India is fragmented, since most departments work independently with-
out links to others. Many bodies do not have planners with backgrounds in the physical,
1004

economic or social sectors, which ultimately dictates the success of policy. The government
needs to incorporate such social aspects into the smart-city program. Smart-cities, even in a
world context, are not about technological solutions, but ultimately about the social services
bettering the quality of life.
2.2 The rural/agricultural background

Agriculture and trade in farm produce is universal, but that does not capture the links to other
spheres of activities; for example, the traditional and historic development associated with land
stems from this base. The land provides not only sustenance in terms of food, or economic sup-
port on the sale of the produce, but also knowledge systems of resources, such as water, and of
time in terms of the passing of the seasons, for planting, harvesting, etc. Also, many cultural
norms in terms of food habits, clothes, shelter, medicine (ayurveda), festivals, etc. were derived
from such rural foundations. It would thus be a limited viewpoint to assume that, in terms of a
pure economic advantage, a wholesale migration would occur to urban areas.
In terms of social indicators, over much of the 20th century (GoI, 2011) birth rates have
decreased from 40/1000 in 1971 down to 20.22/1000 in 2013; the infant mortality rates
have decreased from 165/1000 live births in 1950–55 to 38/1000 in 2015; and average life spans
have increased from 34 years in 1911 to 69 years in 2009. The effective reduction in the number
of farmhands does not necessarily indicate that rural folk feel the need to migrate to cities.
Migration or mobility can be viewed as being caught in the opposing centrifugal/centripetal
forces of the rural–urban divide. It can be highlighted that it is secondary factors that are,
when true, enhancing rural to urban mobility—such as opportunity, education or, to a lesser
extent, lifestyle. It should be mentioned that technology and mechanization has ensured that
fewer hands are required; this would validate some of the changing demographics, but not
necessarily all of the migration. Sociological studies to consolidate these assumptions in the
context of India need to be undertaken more thoroughly, both in rural and urban areas, in
order to comprehend the underlying conditions. Further, it could be necessary to understand
whether or not these migrations are temporary—seasonal or otherwise. It is true that educa-
tion, job skills and similar factors could make people move to urban areas, but it is not fully
known whether or not these are permanent.
Conversely, it can be argued that farmers have not been provided with the necessary level
of institutional support that is required. For example, an input–output related minimum sup-
port price that ensures that the farmer gets the benefit of market prices; a forum that provides
mechanisms for fair trade, regular payment, or is GST (Goods and service tax) enhanced;
institutional mechanisms developed to ensure insurance coverage in case of bad monsoons or
seasonal failures; or award benefits of a bumper crop. With technology, a farmer is allowed
to control the mechanisms of trade, instead of traders and middlemen making much of
the profits. The government needs to develop robust mechanisms for the storage/export of
agricultural items to international markets, or to less developed countries in the form of aid.
It even happens that the government (or representatives), in the name of infrastructure or
project acquisitions, takes over the land of farmers with no compensation, or compensation
that is either minimal or delayed. These kinds of displacements are common, but the govern-
ment often subsequently turns a deaf ear to such occurrences.
The frequent occurrence of farmer suicides are more often than not because they are
unable to pay back the sometimes paltry loans taken for buying seed or fertilizer, which the
government itself could subsidize; however, it is often the case that the government is seen as
ready to write off the loans of the corporate world. Social justice cannot play favorites.
3 THE ROLE OF THE PRIVATE SECTOR
3.1 Land, speculation and investment

Urban growth in India happens without much official involvement. It could seem that the
private sector is ahead of the curve in anticipating change or pre-empting moves by the
1005

government. On the one hand, the government would have a minimum amount of official
information, future plans, and records or processes in place, but much of the expansion
seems to happen without administrative curbs/restrictions. Private individuals are able to
create change, often on a small scale, but sometimes large, without the seeming knowledge of
officials; such changes are often difficult to reverse and are sometimes irreversible.
The expansion of slums in cities is an example that, in apathy to their official responsibili-
ties, as well as finding temporary but real solutions to actual needs, manifests itself in a physi-
cal form. The numbers living in slums was estimated as 40.3 million in the census of 2001.
In 2007, 32.1% of the urban population of India was recorded as living in slums, down from
54.3% in 1990 (UNFPA, 2007).
As against a current worldwide statistic of 50%, the rate of urbanization in India seems rel-
atively slow, having reached 35%. There could be several reasons for this, not all explained. As
mentioned, primary among them could be that the country as a whole stems from that rural/
agricultural base. This has several possible inferences: one is that, with the tradition of farm-
ing, there is an attachment to the land and a knowledge base that is associated with the sea-
sons and nature. Such a tradition dictates much of their lives, in daily, seasonal and lifelong
associations. Although it could sound basic and naive, it is hardly that; this base dictates their
entire life, and it would not be something that they are willing to give up spontaneously for a
job in the city. Therefore, this link is quite strong. (Glaeser et al. 2015), however, wonder why
urbanization in India seems to be hampered—data indicates that mobility in India is much
lower than theoretically expected. Secondly, socio-cultural identity and connections, familial,
caste-based, or historical/traditional, are often binding to the village or to the region. A third
factor, which acts together with the previous one, is the idea of linguistic connectivity; this is
an important part of the decision-making process, especially in relation to interstate mobility
(especially for rural to urban). The limitations caused by this are manifold.
In-state migration makes this easier, especially for those who are not sufficiently literate,
whose linguistic ability is limited to their native language, and this depends on literacy/edu-
cational levels. The average literacy rate in the country is 73%, and this is often a limiting
factor on mobility. A person knowing two or more languages would be comfortable to move
to other states, not necessarily to urban areas within the region, where s/he is confident that
communication will not be an impediment to meeting up with others. Someone who is con-
fident with more than one regional language, as well as another (such as Hindi or English),
would be willing to migrate to many of the metropolitan areas within the country, as long as
it has necessary perks. Intra-state migration is more common for women, possibly because of
family ties or needs, and men are more accepting of interstate mobility.
There are also tertiary factors, such as age, gender, marital status, distance/connectivity
from home, local social networks, etc. The current use of telecommunications and social net-
working has changed this factor substantially. As one advances in age, people are unwilling
to move far from primary social ties, such as family, relations, childhood associations, etc. It
takes a substantial incentive to stay aloof from such social networks without a definite reason.
India has a housing shortage of 18.78 million units (Jha et al., 2015), which is effectively an
annual construction requirement of 2.34 million units per year, as against an ongoing rate of
only 1.2 million units per year. People migrating from the rural areas primarily require afford-
able and basic housing with at least the basic services, which the housing authorities should try
to provide. The country looks to the private sector to fill most of this gap. The national/urban
bodies are unable to take up the role, but may step in using private sector agencies.
4 THE ECONOMIC CITY
4.1 Urbanization and economics

Cities are called the centers for economic progress—in the form of trade, finance or, now to a
lesser extent, manufacturing. Preliminary modern models of growth are centered on industry.
In the early years of development, the coal and steel industries were part of the basic growth
in the eastern parts of the country; mining and heavy industry was centered close to the
1006

location of resources, which could be remote from areas where previous urban expansion had
occurred, and close to transportation networks or trade routes. Mining and similar industries
also meant substantial pollution, and therefore were a lower choice for urbanization.
On the other hand, secondary industries such as spinning mills or goods for daily needs sprang
up after agricultural produce, the primary resource; other trading commodities expanded to
building materials, etc. Classical economists talk about the basics as demand and supply, which
gives value to goods and services, and on which economies prosper or decline. Today’s valuable
products are those of the information age, such as software, networking, etc. Linking these are
infrastructure and networks for transportation, power, telecommunications, etc.
4.2 Urban economics: Feasibility

One hears the complaints of national or urban bodies concerning the lack of necessary funds
for basic infrastructure or its maintenance; it is speculated that the much needed capital in
urban areas can be generated by the proper utilization of resources, such as the land held by
public institutions belonging to the local, state or national bodies. While it is not necessary to
actually sell these assets, other innovative methods, such as renting or leasing to other institu-
tions, public or private, could be a viable method of raising much needed capital (Detter &
Fölster, 2017). The government would then have a financial infrastructure in place that caters
to official needs and need not go to the private sector to raise funds. Many countries have
municipal bonds that are backed by the government to develop such things as infrastructure,
but this has not become common in this country.
For instance, the Indian Railways sits on prime real estate in cities, which can be capital-
ized upon by either building, renting or leasing it to institutions/private enterprises. These
locations fit into the needs of travelers for both affordable accommodation and restaurants,
proximate to their travel plans. The rail stations at locations at Navi Mumbai, planned by
CIDCO, have office spaces that are adjacent, or even directly above, which serve the needs
of offices and of commuters. The railways have ventured into the laying of telecommuni-
cation lines (Railtel) along their tracks, which could provide an essential infrastructure for
the country’s needs. Similarly, the almost defunct post and telegraph (P&T) offices have
not updated their services to include the current requirements of electronic networks or
transportation, instead a new corporate (VSNL/BSNL/MTNL), similar to the private sec-
tor, was developed. Private agencies began taking over their roles, even though their earlier
assets, such as real estate, remain intact. Such symbiotic situations will further the essential
requirements of the country, as well as help to piggyback solutions that are in plain sight
but rarely utilized.
5 THE CASE OF INDIA’S SMART-CITIES
5.1 Technological urbanization

The idea of smart-cities, a recent worldwide phenomenon, has a problem of definition; vari-
ous entities and countries have generated their own to fit their needs. As a technological
development, smart-cities are expected to incorporate e-governance and 24/7 services, such
as superior public transport, water, power and broadband connectivity. The approximated
worldwide market for smart-cities has been pegged at US$ 1.7 trillion.
The government initiated a US$ 15 billion, 100 smart-cities program in 2015, due to an elec-
tion promise made by the new administration; this was a parallel of China’s US$ 300 billion,
193 smart-cities investment (Li et al., 2015). The Chinese government is financially capable of
expanding such a program from the ground up (or greenfield ), because of its established, top-
down decision-making procedures; doing so in India would be expensive. On an experimental
basis, the GIFT-city in Gujarat was begun by the state government, but to get that greenfield
project going has been time-consuming and taken much effort, though there could be lasting
advantages. Instead, the Smart Cities Mission allocated funds for existing cities as part of
their urban renewal process, for the development of these brownfield ones. In addition, other
1007

cities were also helped to build up over several phase- and state-wise competitive allocations
through the AMRUT (Atal mission for urban rejuvenation and transformation) program
under India’s Ministry of Urban Development (MoUD, 2017). The development is planned
for satellite towns, urban renewal and fixes to tier-II/III cities, rather than to build from the
ground up. This also applies to the overall inclusion of the UN’s Sustainable Development
Goals (SDGs).
From the previous term, about 80% of the money allocated for a US$ 20 billion Jawaharlal
Nehru National Urban Renewal Mission (JNNURM), introduced in 2006, was used mainly
in the transportation sector for widening roads and for flyovers; this included many expensive
schemes, such as metros and Bus Rapid Transit systems (BRTs). A review of global cities
along ten indices, such as infrastructure, transparency, long-term thinking, and so on, which
distilled criteria from about 300 factors that was collated from diverse government, research
and academic studies, indicated that none of the cities in the country figured in the top 10/20
listed worldwide, except for a claim of having some of the highest rental values (Feenan et al.,
2017).
An underlying issue is that cities in India never urbanized synchronously with advances
in technology, the provision of infrastructure, governance, or the conversion of economic
assets into on-the-ground development for the public or common good. Theories of spatial
equilibrium (Chauvin et al., 2016) could be more relevant to other regional/global models,
including that of China, but seem inconsistent in the case of India. An answer to Glaeser
(2007), as mentioned earlier, would be differences in language and culture, which would
inhibit mobility. Many such factors, it would seem, do not apply uniformly to developing
countries, especially in the Indian context considering its population scale and social diver-
sity. In many developing countries, social ties to home communities are often robust, which
gives steadiness and support. It was theorized that with progress into market economies with
uniform human capital, perhaps spatial equilibrium could evolve in India. The smart-cities
mission, or its other arms, must also address issues of infrastructure and services, such as
the management of solid waste, use of energy-efficiency and renewables, urban design/plan-
ning, including housing and slums, public–private participation, and e-governance in the
local bodies, and not merely look for solutions to technical issues.
5.2 The case of Dholera

Dholera in Gujarat should be a case study in point regarding smart-city development. Sited
on the Delhi–Mumbai Industrial Corridor (DMIC) and initially a collection of 22 villages of
38,000 inhabitants, it was envisaged by the bureaucrats, without the knowledge or involve-
ment of the inhabitants, to become a showcase for a new ‘Gujarat model of development’,
including the use of blanket, quasi-legal procedures called ‘Special Purpose Vehicles’ (SPVs).
It was expected that this would grow as a greenfield smart-city and by 2040 become larger
than Delhi or Mumbai, if plans were to materialize (Datta, 2015). However, the answer to the
question not addressed at this stage—which was what was to happen to the current and dis-
placed residents, and what was to be their fate—is still unknown. It might follow the dealings
with earlier inhabitants, mainly tribal and rural groups displaced by the Sardar Sarovar Dam,
who have still not been rehabilitated or received compensation after many decades; many of
them were cheated out of their traditional lands by scheming middlemen.
5.3 Alternatives—smart-villages
An alternative to such cities, smart, hi-tech or otherwise, is instead the possibility of smart-
villages. Home-grown solutions, which have been proposed by technologists who want to
see change from the ground up, from the grass-roots, indicate that this development should
happen not merely in selected areas or metropolises of the country, but all over. One such
case is that of PURA (Providing Urban Amenities to Rural Areas), proposed by A.P.J. Abdul
Kalam, an eminent space technologist, and also a past president of the country. He envisaged
that such technological developments should happen from the level of the villages themselves;
1008

with the ongoing development in both hardware and software, the benefits of connectivity
should proportionately flow to the lowest clusters of society—the villages. Essentially, it was
thought that smart-villages need to be developed in order to directly connect the village with
towns and cities throughout the country. It is that knowledge that is at the heart of such
developments, and from knowledge/learning and the understanding of issues and problems,
local or otherwise, technology would only need to be the connector or link within society.
This would probably be an inversion of the current understanding, but the logic of it means
that the two-thirds of the population that still resides in rural areas would be sufficiently
empowered to not need to move to the towns and cities. With the provision of amenities,
(wired/wireless) networks and infrastructure similar to that available in the cities, there would
be no real incentive or need to move.
The arguments are several; on the one hand, economic necessity can be alleviated by the
provision of the forms of income that the technology could bring; technological connectiv-
ity could provide partial answers to rural problems, including that of jobs; self-employment
and innovation could be a partial answer, rather than seeking employment in urban areas;
and solutions to local problems, even relating to society, culture or to epistemological issues,
could be worked out at the rural level itself.
The normal process of a distressed farmer moving to a town to seek employment, tem-
porary or otherwise, who ends up living in a slum with others who have migrated, with no
promise of the basic services of water, power or sanitation, sometimes with no roof or shelter
over their heads, could be minimized.
5.4 Technological tragedies and urbanization

Current forecasts regarding technology indicates automation as being part of this progress.
While this is happening worldwide, being advanced by the developed world, what it implies
has universal implications, even for those countries. Automation is the use of computational
methods for design, production and manufacturing; essentially, much of the work that can
be done by computers will taken over by them and away from human-centered processes,
from accounting to routine assembly or iterative processes that can be programmed. Depend-
ing on the complexity, it could also substitute for surgeons or entertainment. This will have
economic and social implications. Could the educational system prepare the job-seekers with
skills for processes that are still unknown? What are educational systems currently teaching?
Current estimates are that automation could replace most jobs, leaving only 30% of
present jobs intact. From this scenario, if the developmental systems forecast that urbaniza-
tion is happening at an unprecedented rate, then why does the future of urbanization seem so
certain? In India, a priority could be to continue maintaining the upgraded knowledge and
skills of the population in the towns and villages, but also to provide them with facilities that
will give them the advantage of both education-wise, job-wise and future-proofing them. The
knowledge base of the world, especially of traditional systems, of indigenous plants, seasonal
farming methods, and local culture and its arts, still needs to be documented and propagated.
It is difficult to say that urban life will provide the answers to the individual or to a society
that needs to adapt its way of life for an uncertain future.
5.5 A dichotomy of urban intent

If current forecasts regarding technology are indicators, then the country seems to be bank-
ing on the idea that rural–urban migration is this inevitable reality, that the country can
build smart-cities to upgrade them to world-class levels. On the other hand, the plight of
existing cities is of how to accommodate the growing numbers of people. The country antici-
pates a future increase in jobs to be taken up by the increasing number of graduates being
churned out by the rising number of sanctioned (engineering/other) colleges. In contrast,
the modern use of Artificial Intelligence (AI), Machine Intelligence (MI), or automation
dictates that repetitive jobs at the manufacturing level, on the shop-floor or in fields such as
accounting will be reduced by the introduction of automation and robots. Tasks that humans
1009

currently do will be done better by automated systems. Looking at this dilemma, why has the
government not looked into its crystal ball to make definite plans for the future? Considering
the vastness of the issue, smart-cities are only a partial solution.
6 CONCLUSION
The idea that urbanization is inevitable is only preparing the individual for an uncertain
future, especially those from rural backgrounds. Will the move to the city ensure that the indi-
vidual will be able to fight the odds to make a living? Is farming the land an outdated lifetime
concept or pursuit? Are we as a society preparing future generations for an inevitability that
no one knows? The answers are not clear.
Glaeser (2011) examines the city from a post-established perspective and rationalizes
its existence; the McKinsey (2010) perspective is neither a ground up viewpoint nor
multidimensional and proposes the smart-city as a 21st century solution to a socio-economic
inevitability. This approach is fallacious, especially within the Indian context. Other models
of development also need to be looked at concurrently. In the final analysis, smart-cities
could become a reality, but maybe for the wrong reasons—it need not develop for the reasons
mentioned. However, the investment in smart-villages could, it is rationalized, have a greater
benefit to the population. National policies ought to have a multidirectional perspective,
which will drive current thinking and actions for a more definite future.
REFERENCES
Chauvin, J.P., Glaeser, E.L., Ma, Y. & Tobio, K. (2016). What is different about urbanization in rich and
poor countries? Cities in Brazil, China, India and the United States (NBER Working Paper No. 22002).
Cambridge, MA.
Datta, A. (2015). New urban utopias of postcolonial India: ‘Entrepreneurial urbanization’ in Dholera
smart city, Gujarat. Leeds, UK: University of Leeds.
Detter, D. & Fölster, S. (2017). The public wealth of cities: How to unlock hidden assets to boost growth
and prosperity. Brookings Institution Press.
Feenan, R. et al. (2017). Decoding city performance: The universe of city indices 2017. Chicago: Jones
Lang Lasalle Ip, Inc. and The Business of Cities Ltd.
Glaeser, E.L. (2007). The economics approach to cities (Working Paper 13696).
Glaeser, E.L. (2011). The truimph of the city. London, UK: Macmillan.
Glaeser, E.L. (2013). A world of cities: The causes and consequences of urbanization in poorer countries
(NBER Working Paper No. 19745).
GoI. (2011). Census of India. [Online] Retrieved from http://www.censusindia.gov.in/2011census/
[Accessed 25 10 2017].
Jha, A., Sharma, S. & JLL-ASSOCHAM. (2015). Housing for all: Catalyst for development and inclusive
growth. New Delhi: JLL-ASSOCHAM.
Li, Y., Lin, Y. & Geertman, S. (2015). The development of smart cities in China. Cambridge, MA: MIT.
McKinsey Global Institute. (2010). India’s urban awakening: Building inclusive cities, sustaining economic
growth. New Delhi: McKinsey Global Instiute.
MoUD. (2017). http://smartcities.gov.in/content/ [Online] Retrieved from http://smartcities.gov.in/con-
tent/ [Accessed 25 October 2017].
Ravi, S., Tomer, A., Bhatia, A. & Kane, J. (2016). Building smart cities In India: Allahabad, Ajmer, and
Visakhapatnam. Brookings India.
UNFPA. (2007). State of world population. [Online] Retrieved from www.unfpa.org [Accessed 25 10
2017].
1010

Environment management through meditation: A sustainable

approach
Madhura Yadav
School of Architecture and Design, Manipal University, Jaipur, Rajasthan, India
ABSTRACT: The environment of our planet is degrading at an alarming rate. One of

the greatest problems that the world is facing today is that of environmental pollution. The
problem of environmental pollution is addressed by many scholars in different ways. But the
origin of such pollution lies not in the space but in the human mind thus the mind is polluted
first and then the outer environment. Hence until the thought process is cleaned and purified;
all measures of environmental protection will not be addressed in a sustainable way. The
environment is a very vast field. For the purpose of this study only the physical environment
is addressed. Spiritual response is needed to manage the environment. The study will be
based on secondary data. The experiment of meditation on water and farming are studied in
detail to arrive at suitable guidelines for environment management.
Keywords: Environment, Meditation, Environment Management, Water, Agriculture
1 INTRODUCTION
Environment management is not a new concept; the Vedic vision to live in harmony with
the environment was not merely physical but far wider and much more comprehensive. The
Vedic message is clear that the environment belongs to all living beings, so it needs protection
by all, for the welfare of all. The Mahabharata, Ramayana, Vedas, Upanishads, Bhagavad
Gita, Puranas and Smriti comprise the original messages for preservation of the environ-
ment and ecological balance. Nature or Earth, has never been considered a hostile element to
be conquered or dominated. In fact, man is forbidden from exploiting nature. He is taught
to live in harmony with nature and recognize that divinity prevails in all elements, including
plants and animals. The relation of human beings with the environment is very natural as he
cannot live without it. From the very beginning of creation he wants to know about it for self-
protection and benefit. Awareness of the philosophy of karma is fundamental to the working
of the universe. The law of karma (and Newton’s third law of motion) simply states that for
every action, there is an equal and opposite reaction, and that this is true in all parts of the
universe. If we understand this law to be universal and immutable, then we would never think
of harming the earth or its people.
With the ever increasing development by modern man, human activity has interfered with
the fragile and complex interrelationship of our holistic universe and damaged the ecologi-
cal system and natural processes. Humanity has gradually lost its sensitivity to the world
of vibration. We have become more concerned with material development and insensitive
toward nature. Nature is made up of five elements, those being space, air, fire, water, and
earth, which are the foundation of an interconnected web of life, interconnectedness of the
cosmos and the human body. The five great elements (space, air, fire, water, and earth) that
constitute the environment are all derived from Prakriti, the primal energy. Environment
study deals with the analysis of the processes in water, air, land, soil and organisms, which
leads to pollute or degrade the environment. Nature has maintained a status of balance
between and among these constituents or elements and living creatures. A disturbance in
1011

percentage of any constituent of the environment beyond certain limits disturbs the natural
balance and any change in the natural balance causes lots of problems to the living creatures
in the universe. Different constituents of the environment exist with set relationships with
one another. The disequilibrium in nature is due to disequilibrium in man’s mind. However,
the great danger to the world comes from the fact that mankind does not recognize that men-
tal pollution has gained dangerous proportions. Another fact that is generally unknown or is
little realized is that the vibrations from a polluted mind pollute the atmosphere: they make
it tense and vicious. Thus it is mental pollution that has spoiled the whole socio-economic
atmosphere and climate everywhere. We need to reconsider our priorities and values to trans-
form the way we think. Attitudinal change is essential. Dharma—often translated as ‘duty’-
can be reinterpreted to include our responsibility to care for the earth.
Meditation is deep and purposeful thought of eternal truths. It is a method to dwell on
the landscape of your mind with understanding and a means to access your subconscious. It
teaches you to have concentration, single track thinking and instant recognition of deviations
from your intended focus. You learn how to distinguish between thought patterns and to select
those that are positive, useful and lead you to your chosen goals. Meditation restores spiritual
self-awareness and power to the human soul. As we meditate the soul dives in divine power
which is rejuvenating and brings back all your list qualities and inner power. It is a spiritual
law that the practice of meditation is essential to spread positive vibration in the environment.
2 METHODOLOGY
This paper is based on a conceptual framework. It gives emphasis on theory development,

historical research, literature reviews, and critical analyses. It analyses existing theories about
how particular interventions like meditation may work. It establishes a linkage between envi-
ronment and meditation and suggests how environment can be managed in a sustainable way
with meditation being a powerful tool.
At the United Nations conference on Environment and Development held in Rio-de-
Janeiro, Earth summit, the Kyoto Protocol on sustainable development emphasized the key
issues of global environmental concerns. No government at its own level can achieve the goal
of environment protection, until the public participates. Public participation is possible only
when the public is aware of the ecological and environmental issues. The most important and
neglected type of pollution is mental pollution.
2.1 Mental pollution and trickle-down effects

Deep thinking would reveal that most of the present global problems are due to mental pollu-
tion; this pollution destroys a person’s sense of being impartial, considerate and co-operative,
and makes him callous, inimical, violent-prone and vicious. It influences economics, politics,
commerce, business, and all relationships, and destroys real love, kindness and human values.
Stockpiling of nuclear weapons is due to suspicion, fear, rivalry and violent tendencies.
Poverty is due to lack of feeling on the part of rich for the poor. It is also due to the greed of
the wealthy; all kinds of violence is due to anger, hatred or enmity. Thus, all global problems
are in some form of thought pollution or mental pollution.
The remedy to this problem lies in realizing the self and also the relationship with other
human beings. Without a spiritual orientation of the relationship between man and man,
moral and spiritual values cannot be brought into play or be sustained. It is, therefore essen-
tial for man to understand his real and intrinsic nature. If this is not done, it is possible that
science without spirituality would lead mankind to a nuclear catastrophe, population explo-
sion or environmental upheaval. It must be borne in mind that mental pollution is the real
and most dangerous enemy of mankind. Meditation and spiritual wisdom alone can wash
the soul of the thought pollution or what is called ‘mental pollution’.
The global environmental crisis is an external signal calling upon us to address an inner
spiritual crisis. We need or take personal responsibility for lobbing and thinking according
1012

to the natural principal of human values. This is only possible when each of us empowers
ourselves through spiritual study such as meditation.
2.2 Meditation and environment

Meditation is a deep and purposeful contemplation about eternal truths. In meditation practice
human being shift from body consciousness to soul consciousness. Meditation is an effective
tool to realize our self. When we meditate, or we remember God, our thoughts are holy, pure
and unconditional, so we do not create bad, but only holy thoughts and powerful blessings. So
when these powerful blessings and thought power are generated around us and within our envi-
ronment, then it’s raised up. Then other people, who are holy and practicing, also generate the
same thought, the same blessing, and the same power. And like attracts like, and those attract-
ing each other become a very powerful force full of positive energy, which defeats all the evil,
negative influence in this world. That is how the world will become purified and improved day
by day. The positive power generated by spiritual practice can prevent some of the disasters in
the world. We still see disasters happen as if nothing has helped. But, if we did not practice spir-
itually, there would be even more disasters, enough to have destroyed the whole planet long ago.
2.3 The impact of meditation on the environment

The universe consists of five basic elements, these being, earth or land, water, fire, air, and
ether. Each of the five elements represents a state of matter. Earth is not just soil, but it is
everything in nature that is solid. Water is everything that is liquid. Air is everything that is
a gas. Fire is that part of nature that transforms one state of matter into another. For exam-
ple, fire transforms the solid state of water (ice) into liquid water and then into its gaseous
state (steam). Withdrawing fire recreates the solid state. Fire is worshipped in many religious
rituals because it is the means by which we can purify, empower, and control the other states
of matter. Space is the mother of the other elements. The experience of space as luminous
emptiness is the basis of higher spiritual experiences. The effect of mediation on water and
on farming is summarized below.
2.4 Experiments with water

The Japanese researcher Masaru Emoto undertook extensive research of water around the
planet. Mr. Emoto took water samples from around the world, slowly froze them and then
photographed them with a dark field microscope. The following few pictures are of water
crystals from different sources. Generally, clean healthy water creates beautifully formed geo-
metrical crystals, while polluted water is too sick to form any crystals at all.
Water was exposed to different types of music. Distilled water was placed in a bottle
between two speakers and specific music was played for one hour. The water was then frozen
and photos taken. See the following pictures.
Mr. Emoto decided to see how thoughts and words affected the formation of untreated
distilled water crystal. This was done by typing words onto paper, then taping the paper onto
glass bottles overnight.
Figure 1. Microstructure of water molecule from different geographic locations of the world.
1013

Figure 2. Effect of music and dance on water.
‘Thank you’ Love and ‘appreciation ‘You make me sick, I will kill you’
Figure 3. Effect of thoughts on water crystals.
Figure 4. a-b Effect of prayer and written label on water crystals.
It is observed that the negative words ‘You make me sick, I will kill you’ resemble an image
of polluted water. Now we see the effect of thoughts and intent on the water. Below left is
an image of very polluted and toxic water from the Fujiwara Dam. Below right is the same
water after a Buddhist monk (Reverend Kato Hoki, chief priest of Jyuhouin Temple) offered
a prayer over it for one hour. Prayer, that is sound coupled with intention, seems to have an
extraordinary ability to restore the water back to its natural, harmonious, geometric symmetry.
A bowl containing mineral water was placed on the table in front of Dadi Janki, Adminis-
trative Head of Brahma Kumaris World Spiritual University.
1014

With regard to the molecular structure of water, our intent (thoughts), words, ideas and
music have a profound healing or destructive effect on the water, since the human body is
made up of 70% water. Ultimately it means that what we think creates our reality, not just
emotionally but physically. Since the planet is also made up of 70% water these experiments
have profound implications for the environment too.
These experiments are indeed being done in Japan through Mr. Emoto’s work. 300 people,
usually led by someone proficient in focus and meditation will gather around a polluted stream
and focus healing and wellbeing onto the water for approximately one hour. Mr. Emoto has
reported and documented this on film. According to eyewitnesses the polluted water would
become visibly clearer approximately 15 minutes after the meditation.
3 EFFECTS OF MEDITATION ON AGRICULTURE
The Rural Development Wing (RDW) of the Rajyoga Education and Research Foundation
(RERF) is working in partnership with government institutions, NGOs and research insti-
tutes empower thousands of farmers by reconsidering thought-based technology (medita-
tion) combined with organic farming used in rural India for some years.
To face the challenges in agriculture, some farmers decided to undertake experiments on
their farms using Rajyoga Meditation practice in their daily activities.
Farmers using this technique reported considerable improvements in resistance to disease,
pests and adverse weather conditions. The meditation involves creating the awareness of
being the subtle conscious being rather than the physical body and then directing thought
energy (peace, love and power) from the Divine Source to the crops.
Now more than 400 farmers all over India who are practicing sustainable yogic and natural
farming techniques. In India, a number of Agriculture Universities and Researchers have
also taken up the research in order to measure and quantify the actual advantages.
Some preliminary findings have been noted as follows:
Germination rate of meditated seeds was at 93.33% in contrast with non-meditated seeds
which were at a rate of 86.67%. Faster growth rate of meditated seeds, which took six days
less to germinate.
Significant growth in friendly insects’ population
Micronutrient content of meditated crops of wheat showed higher amounts of iron,
increased oil content in oil seeds, improved protein and vitamins in vegetables thus increas-
ing the energy value.
Soil microbial population showed higher population of rhizobium, azotobacter and
azospirllum.
Figure 5. Effect of Rajyoga meditation on water.
1015

Figure 6. Meditation in farming at Mount Abu, India.
Figure 7. Yogic farming at Mount Abu, India.
Other factors, such as root length and seed weight, were greater in meditated samples.
A significant drop in the pest damage was noted. (Results extracted from preliminary report
published by SD Agriculture University).
Basic methodology for the application of positive and elevated conscious thought in farm-
ing, as recommended by the RERF. The seeds are placed in the BK center for up to a period
of one month before sowing. This is then followed by weekly meditations taking place in the
fields by groups of BK teachers and students throughout the crop growth cycle.
Over the last two years, Italy, Greece and South Africa are amongst some of countries
experimenting with these techniques.
4 OVERVIEW OF RESEARCH IN SOUTH AFRICA
This study reviews by Dr. Ndiritu, South Africa of sound based (acoustic frequency) and
thought-based (meditation) technology in improving crop production and assesses their
potential. Published experiences of five farmers report substantial improvements in yield
and resistance to disease, pests and drought. Published data reports yield improvements of
up to 32% and up to 146% in nutritional constituent concentration. The technique is being
promoted non-commercially by an NGO and thousands of farmers have been trained to use
it. The improvements in crop production from acoustic frequency and meditation techniques
are found to be comparable to those from biotechnology. Their potential in mitigating the
global food crisis is considered large. Research and community-based initiatives to promote
the two techniques are therefore recommended.
5 DISCUSSION AND CONCLUSION
The experiment by Dr. Emmoto on water proved that with regard to the molecular structure
of water, our intent (thoughts), words, ideas and music have a profound healing or destructive
effect on the water. The human body is made up of 70% water. Ultimately it means that what
1016

we think creates our reality, not just emotionally but physically. Since the planet is also made
up of 70% water these experiments have profound implications for the environment also.
Brahmakumaris experience of yogic farming substantiates that faster growth rate of
meditated seeds, significant growth in friendly insects’ population, micronutrient content of
meditated crops is higher thus increasing the energy value. Root length and seed weight were
greater in meditated samples. A significant drop in the pest damage was noted.
Meditation training is found to be largely non-profit driven and no additional resources
are required. It is probably the only technology with an effectively negative carbon footprint.
Meditation techniques are adapted for sustainable farming in rural areas to improve crop
production.
Human beings have a greater power of thought than any other living creature. Our thoughts
create subtle vibrations of positive and negative energy and effect our environment to a great
level. Thoughts have power to change the world and therefore it is imperative to understand
that world transformation is through self-transformation.
The process of transformation will only be possible through meditation wherein we are
consciously aware of our true self and our environment. We consciously and subconsciously
create thoughts which are generally not channelized and full of positivity. We do not know
the actions of these thoughts and the environment. Knowledge of generating creative con-
scious and pure thoughts full of energy, positivity and radiating them is possible by the means
of meditation. Change happens from the inside out, from a seed of awareness to a thought,
to an intention and to an action. One of the real benefits of meditation and other reflective
practices is the elevated awareness of the internal and external states of the world.
We have discussed the effects of meditation on water and agriculture. If meditation can
purify the water and crop production in agriculture, why not the physical environment. From
studies it can be concluded that the environment can be managed by meditation.
It is therefore need of an hour to bring it and ourselves to its original state of being the
only way is to get back to silence, restore energy and radiate energy to its zero vibration mode
to stabilize the nature and help itself heal.
Therefore, a mere physical approach is inadequate to manage the environment, but a holis-
tic approach involving both spiritual and physical aspects is required.
The atmosphere we experience is a reflection of our attitude with the experience of the real
power of meditation; we can change deep rooted Sanskaras within the soul and can help oth-
ers and even the elements to change. Create silence within and create a feeling of love for the
elements. Talk to them like service companions. Each element is made of protons neutrons
and electrons that are all moving. Their movement is based on our vibrations. Nature serves
us tirelessly; we also need to serve nature. By spreading peaceful vibrations of peace we can
restore her balance. Let us maintain respect for nature in our hearts and do something to
restore her beauty and balance.
We have to serve five elements of nature to manage environment as below:
AIR is everywhere and gives unconditionally to all living things. It does not withhold its
love from anyone. Instead of polluting air with negative thoughts, to give it the fragrance
of good wishes is to show respect. Ultimately, these reach all living thing through air.
EARTH is stable and nourishing. It gives our feet a place on which to stand. It nourishes
all living things. Just like the earth, our stage should be stable, so that we are not shaken by
anything. Just like the earth, our thoughts and actions should nourish both people and the
environment. On the physical level we have to observe cleanliness.
WATER is flexible and has the power to neutralize. An ocean is a symbol of unlimited
virtues, and like water we should be flexible in dealing with others. We can learn from the
Great Ocean and neutralize negativity. On a physical level even to let the tap drip is to
disregard the importance of water.
FIRE cleanses, purifies and transforms. Fire is a symbol of yoga; our yoga should be intense
like fire so that we are transformed by it. On a physical level to abuse fire is to be irreverent.
Fire rebels against our burning of trees, for example, by filling out air with smoke.
1017

ETHER is subtle yet fat reaching. Ether is so broad that even the plants were thought to
move through it. To become angels our intellect should be like ether—subtle and broad.
To sanctify silence is to respect ether.
The human body is composed of and related to these five elements, and connects each of
the elements to one of the five senses. This bond between our senses and the elements is the
foundation of our human relationship with the natural world. Nature and the environment
are not outside us, not alien or hostile to us. They are an inseparable part of our existence,
and they constitute our very bodies.
All spiritual paths in the world, understand the law of karma to be the incognito mecha-
nism that explains the unfolding of events in the world. The actions we take today write the
script for the circumstances we face tomorrow. If we understand this to be true, then we
might see today’s world as the inevitable result of the choices we made in the past, and in that
instant of understanding, we might see that the only way forward to a peaceful and loving
world would be through our elevated awareness and loving behavior toward our earth and
our human family.
6 FUTURE SUGGESTED RESEARCH
The broader plan is to use the results of these experiments to

Generate enough interest for other research groups at the university to try out ‘mind-
matter’ research
Apply for research funding for field experimentation on yogic farming and analysis of the
impact of meditation on other aspects of water management (governance, dealing with
water-related disasters, climate change etc.).
REFERENCES
Ball State University sustainability webpage. (n.d.). Retrieved from http://www.bsu.edu/ceres/sustainability.

Brimblecombe, P. (1994). Environmental encyclopedia. Air pollution. Detroit: Gale Research Interna-
tional Limited.
Center for energy and environmental sustainability webpage. (n.d.). Retrieved from http://www.cisat.
jmu.edu/cees/.
Deshpande, J. (2007). Meditation Environment through Vipashana Meditation. Mumbai: Daya Publishing
House.
Environment at T. Retrieved from http://www.businessdictionary.com/definition/environment.
html#ixzz2PIVbGDJh.
Henepola, G. (1988). The Jhanas in Theravada Buddhist Meditation (Wheel No. 351/353). Kandy, Sri
Lanka: Buddhist Publication Society. Retrieved from http://www.accesstoinsight.org/lib/authors/
gunaratana/wheel351.html.
Kamalashila (1996, 2003). Meditation: The Buddhist art of tranquility and insight. Birmingham:
Windhorse Publications. Retrieved from http://kamalashila.co.uk/Meditation_Web/index.htm.
Kapleau, P. (1989). The three pillars of Zen: Teaching, practice and enlightenment. NY: Anchor Books.
Mosher, H. (2004). Meditations on the five animals of Tai Chi. Retrieved from Peaceful Warrior Training.
com.
Shashi T. Origin of Environment Science from Vedas. Retrieved from http://www.sanskrit.nic.in/
svimarsha/v2/c17.pdf.
Singh, R.P. (2008). Environmental concerns (The Vedas)—A lesson in ancient Indian history. In Society
and environmental ethics ‘as part of literature published in the International Conference on
Environmental Ethics Education (ICEEE) held on 1617 November, Banmas Hindu University.
Steger, W. & Bowermaster, J. (1990). Saving the earth. New York: Bryon Pries.
Tyler Miller Jr. G. (1987) Living in the Environment. Belmont: Wadsworth Publishing Company.
Vipassana Research Institute (VRI) (n.d.). Bhikkhuvaggo. In The Majjhima Nikaya (second chapter of
the second volume). Retrieved from http://www.tipitaka.org/romn/cscd/s0202m.mul1.xml.
1018

Bicycles for green mobility in urban areas
M. Harisankar
Urban Planning, Government Engineering College, Thrissur, Kerala, India
C.A. Biju
ABSTRACT: The last few years have seen many countries focusing on sustainability in
their developmental agenda. In developed countries the cycling policy has in recent years
evolved from a peripheral matter into one to be considered as a priority, in line with the poli-
cies for other means of transport. Being a clean, green and healthier mode, bicycling holds
the key for sustainable urban mobility and better environmental qualities for our urban areas.
This paper concentrates on the need for urban planners to solve the urban mobility problems
with the introduction and integration of bicycle oriented transit facilities with the existing
transportation network in Indian cities. Planning interventions are very essential in providing
traffic calming and infrastructural support coupled with the appropriate policy backup, so as
to encourage a modal shift from motorized transport to bicycling, which is discussed in detail
in this paper citing some successful examples around the world.
1 INTRODUCTION
The rapid urban growth and increased use of motor vehicles that most countries have expe-
rienced in the recent years has created urban sprawl and higher demand for motorized travel,
leading to a range of environmental, economic and social consequences. This effect is much
pronounced in the case of developing countries. As a sustainable alternative, bicycles can
replace or reduce the usage of automobile transport so there is an urgent need to integrate
bicycles into the transportation system of our cities. This can enhance the mobility of people
in urban areas and can also safeguard the accessibility of our congested cities provided suit-
able facilities are adopted for a bicycle infrastructure. It provides freedom of movement to
rich and poor, young and old alike.
The National Urban Transport Policy (NUTP) has stressed the need for an approach in
transport planning that focusses on people and not vehicles. The united Nations Habitat
Global Report on human settlements also highlighted urban transport with a focus on reduc-
tion in pollution and congestion as a core area for advancing sustainable development in its
Five Year Action Agenda 2012–17 (Planning & Design for Sustainable Urban Mobility).
Indian cities have a latent demand for cycle and walking trips and a topography that suits
it well which can be exploited in the form of strategic planning and designing of suitable
infrastructural support for bicycles, which helps enhanced urban mobility to a great extent.
This study discusses the suitability of bsicycle as a sustainable mode for urban mobility in
urban areas, how far they can be integrated with the existing conditions, their planning and
design considerations and programs and techniques for promoting a bicycle infrastructure.
2 BICYCLES IN AN URBAN SCENARIO
Policies for car oriented transport development have resulted in more and more road
construction that have clearly failed to cope with the ever increasing demand for rapid
motorization that has made our roads more and more congested.
1019

Lewis Mumford in his book The Roaring Traffic Boom sarcastically stated this situation,
‘adding more lanes in highways to accommodate the increasing number of automobiles are
like loosening ones belt as a cure for obesity’. Increasing the lanes in highways and construc-
tion of flyovers can be considered only as short term solutions that can ultimately result in
more vehicles on the road. Long term solutions in solving the mobility problem in urban
centers lies in creating more and more facilities for active modes of transport like bicycles,
thereby reducing the demand for motorized travel.
Cycling offers many benefits to the problem of urban mobility. It is one of the ideal modes
of transportation for trips up to five km. Figure 1 shows the distance/travel time ratio for
different modes of transport. Considering the congestion in our cities, it can be seen that for
up to five km the bicycle is in fact faster than motorized modes of transport.
The dimension of social inclusion is also very important in India when large numbers of
current bicycle users do not cycle out of choice but remain captive as they have no other
viable option. These captive riders are predominantly poor workers and students. Indian cities
have mixed land use patterns and low income households living in urban slums, which are in
close proximity to the planned residential areas. Captive riders use the bicycle for even longer
trips for work (up to 50 km in a day) such is the compulsion of captive riders (Tiwary et al.
2008). Presently the car centric transportation planning in our urban areas has neglected these
groups and have made their mobility even more miserable. Thus, improving conditions for
cyclists can improve the quality of life for poorer groups, those who are less likely to have cars.
Concern about worsening air quality has increased significantly all over the world in the
recent years with the undoubted benefits of cleaner engines and fuels often being undermined
by sheer growth in the volumes of traffic. The transportation sector contributes about 20%
of CO2 emission worldwide and 15% in India (Tiwary et al. 2008) and it is a growing source
Figure 1. Vicious circle of car oriented transport development (Source: Torsten et al 2012).
Figure 2. Distance/ravel time ratio for different transport modes (Source: Torsten et al, 2012).
1020

Figure 3. Trends in bicycle modal share (Source: 2).
of greenhouse gas emissions that contribute to climate change. Cycling provides an excellent
opportunity for individuals to incorporate physical activity into daily life. The human scale
urban environments that support cycling and discourage car use can improve social interac-
tions and increase community attachment, livability and amenity (Litman et al., 2009).
The medium and large cities have a typical bicycle modal share of 13–21% (Figure 3).
Cycle trips might be as low as 6–8% in mega cities, however the absolute numbers are about a
million bicycles. Most of the medium and large cities in India have about 56–72% trips, which
are short trips (less than 5 km in length), offering a huge potential for bicycle use.
3 PLANNING FOR BICYCLES
Planning for cycling involves a combination of infrastructural and non-infrastructural (soft)

measures backed by strong policies to support cycling. For the bicycle to be useful for transpor-
tation, they need adequate route infrastructure (LTN, 2008). Traffic calming makes people more
comfortable on the roads. It can enhance the safety as well as the perceived safety; and it can
also reduce the damage caused by accidents when the traffic speeds are low enough. The concept
of BFZ (Bicycle Friendly Zone) is where a standard bicycle symbol is used in roads or lanes
designed for lower traffic speeds as an important measure in traffic calming. It acts as advice or a
warning to vehicle drivers that cyclists are likely to be ahead and this can also narrow the appear-
ance of the available road space while preserving the space for cyclists (McClintock, 2002).
Making the environment outside the front door into an environment safe for cycling is
the first step in making the choice of cycling a possible one. Another popular measure that
was originally developed in Dutch towns was ‘home zones’/‘woonerf ’ – the living yard. This
is based on the idea that residential streets are planned and reconfigured such that ‘residents
come before cars’. Most of the streets in Amsterdam, the Dutch capital, also known as the
bicycle city of the world, are traffic calmed at 30 km/h, including home zones with a 7 km/h
limit. ‘Bike streets’ are also used here, where bicycles are considered the primary modes of
transport and others are treated as guests. Removing the center line and well-designed road
humps offer speed reduction benefits whilst being more comfortable to the cyclists. Flat
topped road humps can be used as pedestrian/cyclist crossings. A cycle bypass allows the
humps to be avoided altogether. Chicanes are usually constructed using two or more build
outs alternating between each side of the road.
A defined set of bicycle routes makes it possible to travel around a region by bicycle in a
safe and connected manner. In Delhi the BRT (Bus Rapid Transit) corridor was established
and has been operating since April 2008. This corridor is designed to be fully supportive
to NMT users including bicycle lanes and footpaths. Special accommodations for cyclists
at intersections include advance stop lines, bike boxes at intersections, bicycle access lines
and green wave signal timing for cyclists along the route. Network management can help
deliver a pleasant environment for cyclists and can also reduce the need for a cycle specific
1021

Figure 4. Bicycle Friendly Zones (BFZ) in shared roads. (McClinto).
Figure 5. Bicycle tracks in Gurgaon, Haryana (Source: Car-free Gurgaon).
Figure 6. Provision for bike racks in buses (Source: Pucher & Buhler).
infrastructure. This can be done by vehicle restricted areas, redistribution of carriageway

space, self-calming roads where the use of physical features encourages lower speeds and
filtered permeability to certain roads. The control of parking through charges or limiting
capacity and duration of stay can encourage cycling.
4 NON-INFRASTRUCTURAL/SOFT MEASURES
Most of the countries give as much importance to soft measures as they give to infrastruc-
tural measures. Such programs concentrate on the positive sides of cycling. There will be
increased opportunities to promote cycling through sustainable transport schemes such as
“safe-routes-to-school”, “safe-routes-to-leisure-projects” and “bike-to-work-schemes”. Also
1022

it focusses on the promotion of cycling for other purposes other than leisure and work, like
shopping and other utility trips, mainly targeting cycling for women and school children.
Many of the cities in the UK have been framing green commuter plans that assist staff in
managing journeys to and from work in an environmentally sustainable way. Small cities like
Davis in the USA, have been particularly focusing on soft measures. Amsterdam offers bike
training for all children in schools and special bike training programs target groups that cycle
less. Mega cities like London introduced widespread bicycle training in recent years. The city
organizes cycling ‘Ciclovias’ in which part of road network is closed to motorized traffic and
the roads are reserved for bicyclists and other non-motorized modes. The cycling advocacy
group—London Cycling Campaign (LCC) organizes the London Skyride with about 500
local rides and events, as well as promoting cycling training, running repair workshops, and
providing input to hundreds of city traffic schemes with bike route projects.
5 INTEGRATION WITH PUBLIC TRANSPORT
The integration of cycling with public transportation helps cyclists cover trip distances that
are too long to be made by bike alone. Public transportation can also provide convenient
alternatives when cyclists encounter bad weather, difficult topography, gaps in the cycle net-
work and mechanical failures. There are four main approaches: provision of bike parking at
train stations and bus stops, bike racks on buses, permission and storage space to take bikes
on board trains, and the co-ordination of cycle route networks so that paths and lanes lead
to public transportation stops. There are many different kinds of bike parking, ranging from
simple bike racks on sidewalks near public transportation stops to advanced full service bike
stations. There are multi storey bike parking stations being increasingly used in Amsterdam
as an answer to the rising demand of bike parking at important transit stations. The serious
problem of bike theft in most countries has increased the demand for secure parking. In India
such secure bike parking stations need to be set up to increase the cycling levels in our cities.
The overcrowded public transportation system and difficulty in design modifications of trains
and buses makes taking bikes on board trains and buses increasingly challenging.
A public bicycle sharing system has a major role in solving the last mile problems in cities.
It enables the public to pick up the bicycle from any provided self-service docking station,
which can then be returned to the same or any other docking station. The concept of bike
sharing has also been recently introduced in India and few cities have already experimented,
including FreMo in Thane, Green Bike Cycle Rental and Feeder Scheme, Delhi Metro Cycle
Feeder Service in Delhi, Namma cycle in Bangaluru and Cycle Chalao in Mumbai.
Bicycles are an important means of transport in all urban areas in India. At present, most
of the residents in India depend upon non-motorized transport to meet their transportation
needs. The average share of cycles in medium cities of India varies between 3% and 7%. The
years after the industrial revolution showed a massive increase in the number of motor vehi-
cles accompanied with a dip in the average share of bicycles in the major cities of India in
the 1980s and 1990s. A large amount of utility cycling is still present in Indian cities as it is
Figure 7. Trends in bicycle modal share in 1980, 1990 and 2000 (Source: 20).
1023

the most affordable form of transport available to low income households and plays a greater
role in social inclusion. Cyclists in India also face a high risk of getting involved in fatal traffic
accidents. According to Tiwary et al. (2008), cyclists are involved in 5–10% of the total road
related fatalities in medium and large cities. The share of bicycle traffic in many Indian cities
continues to be substantial despite the lack of planned facilities for them.
Considering the international trends some of the cities in India have recently adopted bicycle
infrastructure facilities with their transportation network. In Delhi, the BRT corridor is designed
to be fully supportive to NMT users including bicycle lanes and footpaths. In Lucknow 31 km
of road stretch has already been constructed with cycle tracks. Historically known as the “Cycle
City” of India the city of Pune has about 90 km of cycle tracks. A bicycle master plan is under
construction, which will contain plans for the creation of a city-wide cycle track network and
cycle safe streets, a city wide public bicycle scheme, bicycle parking facilities, integration with
public transport, cycle promotion institutional mechanisms, capacity building and financial
planning for implementing the plan, and awareness and education campaigns.
6 CASE STUDY INFERENCES
Case studies are undertaken based on the method of provision of bicycling facilities in cities of
different sizes. The smaller cities have shorter trip distances, decreased levels of pollution and
less stressful traffic conditions, while large and mega cities with a larger geographical area and
greater trip distances make the provision of such facilities more challenging. Correspondingly,
the census classification of Indian cities is used to match the Indian conditions.
7 ANALYSIS
The major findings from the foreign case studies are compared with the existing condition in
India to examine how far they can be applied in Indian context. In India due to the higher
population and population density compared with the European case study cities, the census
classification is used to divide the cities in India based on size. So further analysis is based on this.
7.1 Traffic calming

In small cities separate segregated facilities decrease the need for traffic calming while in
the large and metropolitan cities strict hierarchy is observed in the provision of bicycling
facilities. Most of the street roads are traffic calmed while the coarser network of segregated
traffic routes are traffic calmed at intersections.
7.2 Network management and bicycle infrastructure

In small cities more importance was given to recreational cycling along the green belt and
segregated cycling facilities owing to greater space availability, lower population density and
better social bonds. In the case of large and metropolitan cities, although a mix of segregated
and shared paths are used, priority was given to getting a continuous network connecting
Table 1. Classification of cities for analysis based on case studies.
European case study Census of India

cities classification Population classification Population
Small cities Up to 300,000 Small and medium Up to 500,000

towns
Large cities and 300,000–1,000,000 Large cities 500,000–1,000,000
mega cities Over 1,000,00 Metropolitan cities Over 1,000,00
1024

major transit points and educational institutions. The network management is carried out
in such a way so as to give an added advantage to cyclists in terms of time and distance.
Measures like congestion charging and travel demand management, along with making CBD
free for motor vehicles can bring positive changes in Indian cities. GIS can also be used in
effective decision making process.
7.3 Soft measures

In the case of small cities, the case study cities irrespective of their size focused on family ori-
ented bike training programs and cycling safety training among children. Imparting cycling
safety awareness is given the prime consideration in devising such policies and programs. The
cities also organize mega events for popularizing bicycling among motorists and bringing
about a modal shift. In Indian conditions it is important to identify the major employers in
the city, promote bike to work programs and commuter subsidies to encourage employees
to use bicycle as a commuting mode. The catchment areas of all schools could be made bike
friendly to encourage more children to use bicycles to commute to school.
7.4 Public transportation integration

In small cities there is less scope for integration with public transport owing to the lower geo-
graphical area and smaller trip distances, they may compete with each other but in the case of
large and mega cities, their larger geographic area and greater trip distances makes it difficult
to cycle the whole distance and necessitates integration with public transit. This can be done
by making provisions for bicycle mounting in public transit vehicles and parking facilities in
transit stations and other areas of commercial activity. In Indian conditions bike racks or
bikes inside public transport buses may not be possible due to overcrowding but a small por-
tion of area in motorized vehicle parking stations can be set aside exclusively as bike parking
stations to provide cyclists a safe and secure place for keeping their bicycles. Where there are
space constraints, multi storey parking stations can be used to increase the efficiency.
7.5 Public bicycle sharing system

The smaller geographical extend and less scattered concentration of transportation terminals
makes a public bicycle sharing system a less attractive option in small cities. However, they
can be integrated with the existing transportation system in large and mega cities to help
improve multi modal connectivity. This is also done by private partnership as their names
can also be included with the name of the bicycle sharing system. However, in Indian condi-
tions, theft and vandalism may prove to be a deterrent factor in providing such facilities. Use
of information technology and provision of safe and secure parking facilities can help to
address such issues. The user fee charged may be free for the first few hours and subsequently
charged, but lower than the costs of public transport.
8 STRATEGIES
Newer cities are likely to have more space for infrastructure, given the wider streets that are
common in these cities. But a lower density, single use development pattern means that des-
tinations are likely to be more dispersed. In such cities, a bicycle infrastructure may be used
more for recreation than transportation, especially if the city builds off-street bicycle paths.
For older cities, more compact, mixed-use development patterns keep destinations within the
cycleable distance but the city might not have enough space for incorporating a bicycle infra-
structure or have greater intensity of vehicular traffic to protect the cyclists. Distances may
make utilitarian cycling feasible in these cities, but creating a perception of cycling is challeng-
ing in these cities. The strategies are given for small and medium towns, and large and met-
ropolitan cities separately along with some common strategies. Although these classifications
1025

may not be strictly followed, as one can see, some of the strategies mix with each other to
provide a general framework for the planning of bicycle facilities in various classes of cities.
Common strategies:
Stress the benefits of cycling to the individual and society with community participation at
all stages of planning and development
Develop a bicycle masterplan for urban areas and periodic updates
Carry out controversial decisions in stages
Combine incentives for cycling with distinctive car use
Nurture coalitions with bicycle groups, city administration, politicians, other bike-
friendly groups, national organizations and technical people right at the beginning of
implementation
Strategies for small and medium towns:
Promotion of cycling among school children by expanding bike-to-school programs and
cyclist education
Creating safe cycling conditions by physical segregation from high speed motor vehicles
Planning for greenways around the outskirts of the city
Strategies for large and metropolitan cities:
Build a network of integrated bikeways with intersections that facilitate cycling with spe-
cial consideration of the areas of low income households
Integration with public transportation and implementation of bike sharing programs
Physical segregation of fast moving and slow moving vehicles along BRT corridors
Restrict car use through traffic calming, car free zones and less parking areas and mainte-
nance of bike lanes already provided
9 CONCLUSION
The extent to which the bicycling facilities can be integrated with the existing scenarios of the cit-
ies depends on the city size, its inherent qualities and the existing transportation facilities in the
cities. Indian cities truly reflect the large potential for bicycle use with a huge latent demand for
bicycle facilities and infrastructure. The mixed land use and poly nucleated city structures with
a higher percentage of short trips gives an added advantage to providing bicycle facilities while
the large percentage of informal housing, high population density and lack of space presents
unique challenges. At the same time, there is a great need to provide such facilities in Indian
scenarios. So, the planning for bicycle facilities cannot be carried out in isolation but in an inte-
grated approach, addressing the needs of all road users. It is a combination of various infrastruc-
tural measures coupled with policies and promotional support programs, backed up with strong
political will and community participation. Cycling offers a healthy, cost effective, equitable way
to improve the sustainability of urban transportation systems and build more livable cities.
REFERENCES
Belter, T., von Harten, M. & Sorof, S. (2012). Costs and benefits of cycling. European Union.
Cycling infrastructure design—how bicycling can save the economy, TSO publishing house.
McClintock, H. (2002). Planning for cycling. Cambridge, England: Woodhead Publishing in Environ-
ment Management.
Pucher, J. & Buehler, R. (2012), City cycling. Cambridge, Massachusetts: The MIT Press.
Rastogi, R. (2011). Promotion of non-motorized modes as a sustainable transportation option: Policy
and planning issues. Current Science, 100(9), 1,340–1,347.
Tiwary, G., Arora, A. & Himani Jain, H. (2008). Bicycling in Asia. TRIPP, IIT Delhi.
1026

Impact of foreign direct investment on the city: Form and growth

of cities
Amal Mathew & Manoj Kumar Kini

Department of Architecture, College of Engineering, Trivandrum, Kerala, India
ABSTRACT: The economic policies introduced by the government of India in 1991 played
a great role in shaping the cities of modern India. The flow of revenue from foreign countries
resulted in a major change in city structure. Availability of basic infrastructure was the major
determinant for the inflow of Foreign Direct Investment (FDI) which influenced the states to
improve the facilities. The Multi-National Corporation (MNC) investment region witnessed
large-scale development resulting in large demographic changes due to migration. Booming
land prices in the area forced growth in a vertical direction, which has impacted the city form.
The biggest challenge in FDI-induced urbanization is making the city inclusive. Developments
should be made with a vision for the future. A spatio-economic development policy wherein the
spatial structure dictates the growth of the economy by attracting the investments can dictate
the growth of the city by enhancing the positive and reducing the negative impacts of FDI.
1 INTRODUCTION
1.1 Foreign direct investment

Foreign Direct Investment (FDI) or foreign investment refers to the net inflows of investment
to acquire a lasting management interest (10% or more) in an enterprise operating in an
economy other than that of the investor.
FDI has turned out to be a major economic policy of the world that ultimately had a lot
of spatial relations. The growth pattern in the host city varies with the type of investments it
procures. Foreign investments bring with them new trends and techniques in construction,
market structure and quality of life and lifestyle. FDI has resulted in the creation of polycen-
tric cities. As India is one of the top countries to attract foreign investment, a case study like
this could help in understanding and formulating guidelines for the development of a city in
receipt of FDI. FDI brings about many impacts in various areas such as social, economic
and physical. This study is limited to the impacts in physical form whereby the changes in the
city’s structure and form is studied.
2 IMPACTS OF FDI
2.1 Economic impacts

With the influx of foreign investment, the economy of a region can elevate to a higher level.
The companies that need a lot of skilled labor will provide a good salary to their workers
whereas unskilled labor will be negatively affected as it will be so easily available at their
disposal. This would again result in the widening of the gap between the rich and the poor.
The people who best benefit from these developments are the middle class who can enjoy a
better buying power as the Multi-National Corporations (MNCs) bring in a lot of offers and
discounts in order to capture the local market. The local players in the same industry will be
1027

affected and they will have to come up with measures to compete with these companies. The
scenario of branding in the organized retail sector happened in this regard.
2.2 Social impacts

The MNC investments in the region cause the formation of a new middle class/high-level pro-
fessional workforce that occupies white-collar jobs and is characterized by a shared educa-
tion, lifestyle, consumption patterns and occupations. Social polarization in society becomes
stronger as changes in the absolute and relative size of different income groups develops over
time—directly inferred through the structure of the high-income lifestyles of those employed
in the service sector (e.g. high-income residential and commercial gentrification). Infrastruc-
ture and streetscape are developed to match world class standards to attract a safe and stand-
ardized environment. To provide safe and secure environment to the MNC, the security of
the areas are improved and is the first consideration of the governing authority.
2.3 Physical impacts

Emergence of new retail and wholesale: New buildings and building typologies are becoming
common (e.g. malls, hypermarkets). In order to compete with upcoming foreign brands and
retail chains the big, native retailers have also stepped into branding and started retail chains,
for example, Spencer’s, More, Reliance Fresh, and Unlimited Stores.
Huge warehouse-type blocks are introduced in the cities to house multi-brand retail and
wholesale stores such as Walmart and Metro. Isolated knowledge campus with various
relaxation policies are publicized and supported like Special Economic Zones (SEZs). The
development pattern is mainly focused around the area with MNC investments like IT or
industrial hubs, with the emergence of new housing typologies (service apartments) and iso-
lated neighbourhoods (fragmented development pattern).
Emergence of cloned built spaces: To attract the new middle class of the city as well as the
foreign visitors, cloned spaces are developed in the city, for example, McDonald’s, Domino’s,
KFC, Levi’s, Zara, Audi, and BMW.
The physical impacts of FDI can be broadly classified into macro and micro levels, wherein
the macro level affects the whole structure and functioning of the city and the micro level
impacts predominantly the built form characteristics. The impacts caused in these areas form
the crux of this study and formulate the guidelines needed for enhancing the positives and
reducing the negatives, which is also a concern.
3 MACRO-LEVEL PHYSICAL IMPACTS
3.1 Emergence of polycentric cities

Formation of polynuclear cities as a result of new developments which require large parcels
of land. Growth of cities toward the periphery and around major junctions where the new
developments take place. Formation of squatter settlements around the newly developed
regions to cater the poor.
Figure 1. Polycentric city. Figure 2. Formation of squatter settlements.
1028

3.2 Agglomeration economy
Companies will be interested in investing in areas where similar types of industry are present.
A region within a city often becomes the hub of a particular type of industry.
3.3 Changes in land use

As the demand for industrial use increases, the land available will be under pressure and con-
version to industrial and commercial land use takes place. High increases in land values pave
the way for the creation of luxury gated communities. Unaffordability of housing stock in the
city forces the working class to migrate toward the periphery and outskirts.
3.4 Grain and texture

The new building typologies which occupy large footprints result in a change from fine grain
to coarse grain. Planned areas like townships induce a new texture.
The skyline of the city changes with huge structures now forming the backdrop of the city.
Figure 3. Agglomeration economy. Figure 4. Industrial agglomeration.
Figure 5. Land use changes. Figure 6. Shift towards the outskirts.
Figure 7. Change in the grain.
Figure 8. Change in skyline.
1029

4 MICRO-LEVEL PHYSICAL IMPACTS
4.1 Cloned typology

The establishment of international franchises who follow the same design language every-
where, irrespective of the location, has resulted in the creation of buildings that are cloned
from a foreign setting.
Foreign franchises will attract local customers and the local players will be forced to
improve their standards to match the foreign players, which helps in improving the overall
quality and competitiveness of the local players. This results in an increase in the cost of the
products, which in turn increases the cost of living.
4.2 Warehouse typology

Retail and wholesale markets become dominated by the foreign players and they offer a wide
variety of products within their showrooms, which are like large warehouses. Large retail and
wholesale shops become a hub of activity or a node. Large local players will try to imitate the
foreign players, while the smaller players get eliminated.
This will provide the middle class with better buying power and better products. Heavy
traffic congestion in areas housing this type of hypermarket is a common phenomenon.
4.3 Gated communities

Large plots of lands become gated communities containing all sorts of amenities for the resi-
dents. As the number of gated communities increases (due to the increase in housing needs),
this has a huge impact on the skyline of the city.
These gated communities will be located at prime locations and hence the prices will be
very high, which makes housing unaffordable to the lower middle class forcing them to move
to the outskirts.
Figure 9. Cloned typology.
Figure 10. Warehouse typology.
1030

4.4 Shopping malls
Shopping malls emerge as a result of local market agglomeration. A large impact on the
city structure occurs as the shopping malls become the anchors and are usually placed at a
strategic location.
Because the city lacks parks due to the very high land prices, malls become a place of
recreation. All this contributes toward the traffic congestion in areas nearby, also influencing
the movement pattern.
4.5 Office buildings

Major players such as IT businesses require large office spaces, resulting in huge structures,
often with a glass façade. The city’s skyline is highly affected by these massive structures,
which are situated in dominant positions in the city.
Figure 11. Gated community.
Figure 12. Shopping malls.
Figure 13. Office building.
1031

Increases in the temperature of a city with many reflective glass façades, along with the high
energy consumption of air conditioning within them, are causes for concern with these buildings.
The emergence of new typologies in the built fabric as a result of FDI brings in changes in
the image ability of the city. The buildings that spring up reflect global standards, making the
city global. New typologies such as cloned buildings, warehouse typology, office buildings,
shopping malls and gated communities occur as an impact of FDI.
5 GUIDELINES FOR FUTURE DEVELOPMENT
5.1 Need for regionally global developments

The major impact on the growth and built fabric of the city is the formation of global cities.
The Indian city, with its inclusive society whereby the various classes of people live in an
inclusive city, is being affected by the very idea of the global city. The city becomes a place for
the rich and the well-to-do. This can be understood through the large morphological changes
that are happening in the city.
An area where a SEZ is proposed, in order to attract FDI, goes through a sudden change.
Speculative investments occur in that region so as to reap the benefits in the future. Once
the development starts the change will be drastic. A region with fine grain and texture gets
converted into a region of coarse grain with the building of large structures such as office
buildings and malls, which are part and parcel of any foreign investment.
The so-called global city structures, which are basically glass towers, change the skyline
and character of the region. The traditional and regional style of architecture is completely
forgotten and an alien setting is created in the name of the global city. No mandate is given to
the investors to come up with glass towers; thus buildings responding to the regional climate
and setting need to be proposed in these areas in order to make the area regionally global.
5.2 Open space

When an investment comes into an area, the land value shoots up, which makes people want
to build structures wherever possible. This phenomenon leads to the major issue of lack of
open space. Cities which grew without a proper master plan witnessed such a growth. Proper
measures should be taken so that the open spaces needed for the development are defined in
the master plan stage. This lack of open space is clearly visible in Gurgaon, where no consid-
eration for open space was given in the initial stages.
5.3 Envisioned connectivity

In an existing city, new developments might be on the outskirts, the periphery or in the
suburbs and these developments in turn become the anchor functions which make the city
Figure 14. Connectivity to various city centres.
1032

polynuclear. Envisioning this type of growth at the time of inception of the master plan
should be a given so that, once they develop, connectivity with the initial core should not be
an issue. Traffic congestion in the newer developments like Whitefield in Bangalore can be
accredited to this scenario.
5.4 Squatter settlements

Formation of squatter settlements is a common phenomenon in all emerging cities. When a
new master plan is prepared there should be a mandate to allocate affordable housing units
for the poor so as to avoid the formation of these types of settlements. All cities have this
problem, which needs to be addressed.
5.5 Farmland
Any development should address the ecological sensitivity of the region; productive farm-
land should not be acquired for the development process. The agricultural production and
livelihoods of many have been lost in the past with the setting up of SEZs and other devel-
opments. First preference in development should be given to unfertile land so that farming
activity is not lost, and the fertile land continues to be farmland. The past experience of
Nandigram in West Bengal points toward this aspect.
5.6 Security
Security issues are a major concern of these newly developed cities. In the case of Gurgaon,
the city faces lots of security issues as any nightlife is present only within the glass
buildings. The segregation and planning made the streets dead at night. This scenario,
along with the gated communities that act as another entity completely secluded from the
happenings outside their gates, increases security issues. Events outside these compounds
are unknown to their residents, making the streets unsafe at night. Introducing various
functions at the same place at various point of times of day is an option in tackling
this situation. Enhancement of the night life in the cities is an important aspect to be
considered.
5.7 Anchor points in a polynuclear city

When a development such as a shopping mall occurs, it becomes an anchor point in the
city. This happens because developments like these become a hub of activity and recrea-
tion for the city dwellers because the city lacks open spaces and public recreation facilities.
This scenario introduces many other impacts to the city because the traffic structure and
the movement pattern are influenced as these structures become anchor functions. Such
strategic developments should be allowed only in areas where the city’s movement will not
be hindered.
Figure 15. Poly nuclear city with its anchor points.
1033

6 CONCLUSION
The spatial manifestations of FDI are so evident that it can dictate the way in which a city
develops. FDI-induced development can be seen all around the world wherever the policy of
FDI is enacted. The growth of Chinese cities is the best example in this regard.
The biggest challenge in FDI-induced urbanization is making the city inclusive. When the
Indian city, with its chaos and disorder, is converted into a global one or a new development
such as SEZ, the low-income groups are not given much consideration. FDI developments
bring a lot of employment opportunities especially the white-collar jobs. The job options for
the poor and the ones affected by land acquisition are limited to jobs such as security guards,
and gardeners, making the division in society more pronounced. When the city develops over
a certain limit, it becomes a place for the rich only. Even the middle class finds it difficult to
afford housing in the cities.
The major factor that should be considered is that the developments should be planned.
Developments should be made with a vision for the future. Traffic congestion in many cities,
when the city expands beyond borders, is due to the lack of vision in the early planning stage.
There is a need for a spatio-economic development policy whereby the spatial structure
dictates the growth of economy and the attraction of investment. Western countries have
started developing in this manner because the shortcomings of the older developments are
becoming more and more visible. Thus, to conclude, FDI has a large impact on the urban
form and structure with a number of positives and negatives. It is hoped that a proper spatio-
economic policy to guide the same can enhance the positives and reduce the negatives.
REFERENCES
Chadchan, J. & Shankar, R. (2009). Emerging urban development issues in the context of globalization.
Journal of ITPI (Institute of Town Planners India), 6(2), 78–85.
Chen, Y. (2009). Agglomeration and location of foreign direct investment: The case of China. China
Economic Review, 20(3), 549–557. doi:10.1016/j.chieco.2009.03.005.
Eldemery, I.M. (2009). Globalization challenges in architecture. Journal of Architectural and Planning
Research, 26(4), 343–354.
Mathur, O.P. (2005). Impact of globalization on cities and city-related policies in India. In H.W.
Richardson, C-H.C. Bae (Eds.), Globalization and urban development (Advances in spatial science).
Berlin, Germany: Springer-Verlag.
Mukherjee, A. (2011). Regional inequality in foreign direct investment flows to India: The problem and
the prospects. Reserve Bank of India Occasional Papers, 32(2), 99–127.
Shivam, N. & Keskar, Y.M. (2014). Impact of foreign direct investment over the city form, a case of
Hyderabad City, India. International Journal of Innovative Research in Science, Engineering and
Technology, 3(2), 9674–9682.
Wei, W. (2005). China and India: Any difference in their FDI performances? Journal of Asian Economics,
16(4), 719–736. doi:10.1016/j.asieco.2005.06.004.
Zhu, J. (2002). Industrial globalisation and its impact on Singapore’s industrial landscape. Habitat
International, 26(2), 177–190. doi:10.1016/S0197-3975(01)00042-X.
1034

Author index
Abdul Samad, P.A. 493 Balasubramanya Murthy, Divya, A.H. 611

Adarsh, S. 79 K.N. 857, 865 Divya, D. 277
Ahmad Rather, M. 657, 663 Banerjee, S. 851
AjalBabu, K. 687 Basavaraja, J.S. 381 Elangovan, S. 423
Ajayakumar, R. 721 Basheer, N. 987 Elias, J. 501
Ajeesh, K.N. 617 Bawane, O.P. 943
Ajithkumar, K. 357 Beena, K.S. 271 Farhana, F. 299
Akhila, M. 235 Bharadwaj, A. 857 Fathima, R.K. 743
Akhtar, S. 29 Bhaskar, S. 835 Fathima, S.A. 751
Akkara, J. 149 Bhaskaran, S.K. 845
Amritha Varsha, T.S. 293 Bhat, S. 865 Garg, H. 865
Anagha, K. 271 Bhavyalakshmi, A. 811 Gayathri, A. 671
Anand, N. 531 Biju, C.A. 965, 1019 Gayathri, G.S. 299
Anctil, A. 213 Bimalkumar, M. 401 Gayathri, S. 229
Anilkumar, P.S. 651 Bindhu, B.K. 205 Geeva, G. 179
Anjali, E.P. 799 Bindu, C.A. 973, 987 George, A.M. 549
Anjana, R. 575, 603 Bindu, J. 223 Giri, P. 829
Anjaneyulu, M.V.L.R. 135, Bindu, S.S. 469 Gokul, V. 307
143 Binia, R.K. 67 Gopalakrishnan, M. 925
Antony, A.A. 651 Binu, R. 779 Gopi, S. 451
Antony Raja, S. 623 Bisna, N.D. 805 Gopi, V.P. 701
Anudeep, Y.V. 555, 591 Boban, J. 365 Gopinathan, V.T. 307
Anulal, S.S. 727 Brijmohan, K. 701 Gopu, C. 459
Anusha Nair, B. 265 Gouri, K.S.R. 51
Anusha, C.K. 687 Chand, D.S. 423 Gowri, D.R. 323
Anusree, T. 879 Chandni, S. 265 Gupta, A. 865
Aravind, S.P. 543 Chandrakaran, S. 253
Arjun, O. 435 Chandran, R. 335 Harikrishna, M. 135, 143
Arun, N.R. 243 Chandran, S. 799, 889 Harisankar, M. 1019
Arun, S. 43 Cherian, R.M. 101 Helen, K.J. 895
Asha, N.P. 299 Hridya, V. 179
Ashok Babu, T.P. 443 Daniel, C.B. 157 Hukkim Raja, A. 357
Asna, V.K. 873 Daniyal, M. 29 Hussain, Z. 643
Aswathy Nair, S. 243 Das, G. 229
Athira, H.S. 299 Deepa, R.S. 3 Indulekha, K.P. 87, 109
Athira, K. 701 Deepak, G.P. 393 Irshad, M. 785
Athira, K.S. 351 Deepthi Chandran, R. 293
Athira, S. 287 Devadas, V. 957 Jacob, A. 149
Athira, V.S. 299 Dey, J. 851 James, A. 799, 889
Austin, A. 59 Dhanasekaran, S. 681 Janishali, K.K. 451
Azam, A. 29 Dhanesh, L. 109 Jaya, V. 287
Dhanya, P. 95 Jayakumar, J. 995
Balamanikandan, D. 357 Dhrisya, V. 883 Jayakumar, V. 329
Balan, K. 243 Divya, A. 185 Jayakumaran Nair, A. 549
1035
Author Index.indd 1035 7/9/2018 3:34:10 PM

Jayamohan, J. 243, 249, Manoj Kumar, K.V. 825 Raj, A. 293
265, 293 Manoj, N. 583 Rajan, R. 715, 721
Jayaraj, J. 259 Mansoor, C.M. 825 Rajesh, R. 205, 323
Jayasree, M. 811, 817 Mathew, A. 1027 Rajesh, R. 835
Jayasree, P.K. 277 Meera, C.M. 7 Ralphin Rose, J.B. 293
Jeenu, G. 13 Megha, C. 565 Ralphin Rose, J.B. 307
Jeenu, M. 293 Meghavathu, D. 643 Ramachandran, K.K. 371
Jenifer, L. 603 Meher, P.A. 51 Ramesh Krishnan, S. 335
Jerard, L. 43 Melitt Akhil, P. 793 Ramesh, H. 185
Jinesh, K.J. 149 Menon, D. 51 Ranganayakulu, R. 511
Jiniraj, R.B. 277 Midhun, V.C. 485 Raseek, C. 743, 751, 757
Jithesh, P.K. 485 Mohammed, J. 479 Raseena, A.K. 933
John, A. 933 Mohit Kumar, C. 631 Ravi, J. 727
Jose, A.K. 229 Muralidharan, V. 401 Ravikumar, S. 401
Joseph, G. 21 Reby Roy, K.E. 469, 479
Joy, D.P. 707 Nabeel, N. 933 Rejini, V.O. 611, 651
Joy, J. 817 Nair Radhika, P. 299 Remjish, R.S. 165
Jyothir Rose, K.J. 603 Nair, L.P. 229 Renju, D.R. 243
Nair, R.S. 611 Reshma Babu, S. 299
Kafle, S. 469 Narayan, B. 995 Reshma, R.S. 715
Kala, M.T. 765 Natarajan, S. 857, 865 Riyas, K.K. 701
Kamath, D. 213 Naveen Rajeev, R. 315 Rone Maria, A. 895
Kamble, S.P. 565 Neena, M.J. 135 Roy, A. 873
Kanagasabapathy, S. 401 Ng, K.W. 351
Kannah, D.L.V.V. 423 Nihal Pandey, N. 191 Sachithra, K. 565
Kannan, K. 229 Nijila, M. 765 Sadique, A. 371
Kathiresan, V. 357 Nikhil, M.C. 435 Sahadev, S. 637
Kedia, A. 531 Niloofar, M.U. 87 Sajeenabeevi, B. 623
Keerthi, K.R. 223 Nimisha, N. 873 Saji, A.M. 249
Khader, A.A. 259 Nishan Ahammed, V. 537 Sajikumar, N. 101, 117, 123,
Krishna, R.S. 293 Nithin, S. 493 129
Krishnapriya, S. 575 Nithin, V.P. 493 Sajith Babu, C. 451
Krishnarchana, K. 583 Nived, N. 493 Sajith, P. 265
Kumar Kini, M. 1027 Salim, S. 617
Kumar, D. 531 Pallan, A.P. 623 Sam, A.D. 3
Kumar, V. 637 Pandey, S.K. 851 Sambrekar, A.A. 341
Kumaraswamy, M.B. 381 Paras, 829 Sandeep, K. 417
Kunju, N. 687 Parvathi, S. 493 Sandeep, M.N. 271
Parvathy, S. 287 Sandhini, S. 779
Laiju, A.R. 197 Philip, G.J. 469 Sangeetha, M. 143
Lakshmi, A. 951 Prabhakaran, P. 21 Sangita Dash, S. 583
Lal, D. 253 Prabhakaran, P. 951 Sanil Kumar, A. 957
Latha, K. 555 Prakash, S. 411 Sanjeevy, C. 501
Lawrence, A. 365 Prasanth Kumar, K.P. 537 Sankar, N. 235, 253
Lekshmi, C. 243 Praseeja, A.V. 123 Saravanan, R. 315
Lincy, C.Y. 637 Priya, K.L. 79 Sarkar, S. 197
Priyadarsini, R.S. 59 Sekar, M. 417
Madhan Kumar, A. 357 Priyaja, K.V. 707 Sekkar, V. 637
Madhusoodanan, K.N. 695 Punneliparambil, A.R.G. Seshagiri Rao, A. 511
Manesh, K.K. 365 805 Shaheem, S. 165, 171
Manilal, A.M. 523 Shaheer, P.V. 371
Manju, M.S. 537 Radhakrishnan, K.B. 549 Shaik, S.V. 443
Manju, V.S. 157 Rafeeque, P.C. 845 Shaji, K. 393, 485
Manjusha, J. 889 Rahamathulla, K. 793, 879 Shaju, A. 681
Manjusha, P.D. 757 Rahul, M. 773 Shalimol, V.P. 979
1036

Shanthini, K.S. 707 Sujith Kumar, R. 617 Varghese, L. 365
Shari, E.K.R. 129 Sujith, K.M. 979 Varghese, P. 1003
Sharma, V. 191 Sujix, S. 913 Varma, C.G. 623
Shine, S. 773 Sumam, K.S. 101, 117 Vasanth, D. 631
Shrihari, S. 185 Suman, P. 643 Veeramanikandan, R. 307
Shyam Praveen, P. 479 Sunil Kumar, C.P. 459, 493 Veluru, S. 643
Silpa, K.S. 785 Sunny, G. 249 Venkat Rao, P. 555
Singh, R. 829 Suresh, R. 965 Venkateswarlu, P. 591
Sinith, M.S. 671 Suresh, V.R. 249 Venkatramasubramanian,
Sminesh, C.N. 873, 901 Surya, S. 925 V. 315
Smitha Alex, A. 637 Swamy, K.R. 235 Vignesh, V. 315
Smitha, K.K. 67, 411 Swaraj, K.P. 907 Vijaya, N. 951
Soloman, P.A. 523, 543 Swetha, T. 701 Vijayan, A. 265
Soman, M. 37 Vijayan, S. 265
Soorya, S. 79 Thampi, M.S. 243 Vijayasree, B. 37
Soorya, S.R. 259 Thanigaiarasu, S. 423 Vikramaraj, M. 523
Sreedhu Potty, P.S. 293 Thara, S. 117 Vinay, A. 857, 865
Sreejish, A.G. 901 Thasleema, T.M. 835 Vincent, M. 733
Sreejith, B. 435 Therattil, J.P. 149 Vinod, P. 229, 281
Sreelakshmi, A. 13 Thomas, A.A. 249 Vipin Kumar, K.S. 883
Sreelakshmi, P.V. 671 Thomas, R. 95 Vishnu, C.R. 341
Sreelekshmi, B. 695 Tukaram Bai, M. 555, 591 Vishnu, R. 727
Sreelekshmi, S. 171
Sreelekshmy Pillai, G. 281 Uday Bhaskar Babu, G. 511 Wilson, G. 701
Sreeshma, K. 907 Udayashankar, H.N. 191
Sridharan, R. 341 Usman Abdulla, S. 973 Yadav, M. 1011
Srijith, R.S. 401 Yoosaf, P.P. 459
Srinivasan, A. 857 Vaideeswaran, R. 315
Subha, V. 7 Vaishnavi, L.K. 243
Suhana, N. 323 Vanchipura, R. 329
1037


Webpdf

Uploaded by

Copyright:

Available Formats

Webpdf

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Webpdf

Uploaded by

Copyright:

Available Formats

EMERGING TRENDS IN ENGINEERING, SCIENCE AND TECHNOLOGY

FOR SOCIETY, ENERGY AND ENVIRONMENT

ICETEST_FM.indd 1 7/9/2018 4:25:53 PM

Emerging Trends in Engineering,

ICETEST_FM.indd 3 7/9/2018 4:25:53 PM

© 2018 Taylor & Francis Group, London, UK

Typeset by V Publishing Solutions Pvt Ltd., Chennai, India

Published by: CRC Press/Balkema

ISBN: 978-0-8153-5760-5 (Hardback)

ICETEST_FM.indd 4 7/9/2018 4:25:53 PM

Civil Engineering Advancements for Sustainable Infrastructure

Hydraulics and water resources engineering

ICETEST_FM.indd 5 7/9/2018 4:25:53 PM

Traffic and pavement engineering

ICETEST_FM.indd 6 7/9/2018 4:25:53 PM

Progressive Developments in Mechanical Engineering (PDME)

ICETEST_FM.indd 7 7/9/2018 4:25:53 PM

Industrial engineering and management

Manufacturing technology and material science

Mechanical design, vibration and tribology

Numerical models and computational methods

Other topics related to mechanical engineering

ICETEST_FM.indd 8 7/9/2018 4:25:53 PM

Thermal sciences and transport phenomena

International Conference on Advances in Chemical Engineering (ICAChE)

ICETEST_FM.indd 9 7/9/2018 4:25:53 PM

Electronics, Signal Processing and Communication

ICETEST_FM.indd 10 7/9/2018 4:25:53 PM

Non-destructive classification of watermelon ripeness using acoustic cues 715

Predominant instrument recognition from polyphonic music using

A technique for countering the integer boundary spurs in fractional phase-locked

Adaptive neuro-fuzzy based base station and relay station deployment

International Conference on Emerging Trends in Computer

A review of extractive sentence compression 751

ICETEST_FM.indd 11 7/9/2018 4:25:53 PM

International Conference on Changing Cities-Architecture and Energy

ICETEST_FM.indd 12 7/9/2018 4:25:53 PM

Policy and assessment scenario

Smart city initiatives

Author index 1035

ICETEST_FM.indd 13 7/9/2018 4:25:53 PM

ICETEST_FM.indd 15 7/9/2018 4:25:53 PM

Plenary sessions and keynote speeches

ICETEST_FM.indd 17 7/9/2018 4:25:53 PM

Conference secretariat and committees

Prof. C. Raveendranath—Hon’ble Minister for Education, Govt. of Kerala

Dr. K.P. Indiradevi—Director of Technical Education, Govt. of Kerala

Dr. Jiji K.S.—Associate Professor, Electrical Engineering, GEC TCR

Dr. Anjana R.—Asst. Professor, Chemical Engineering, GEC TCR

Prof. K.R. Rahul—Asst. Professor, Mechanical Engineering, GEC TCR

Dr. E.A. Subaida—Dept. of Civil Engg.

INTERNATIONAL ADVISORY COMMITTEE

Dr. Lars-Erik Lindgren—Professor, Department of Engineering Sciences and Mathematics,

ICETEST_FM.indd 19 7/9/2018 4:25:53 PM

NATIONAL ADVISORY COMMITTEE

Dr. Devdas Menon—Professor in Structural Engineering, IIT, Madras

ICETEST_FM.indd 20 7/9/2018 4:25:53 PM

Dr. R. Rudramoorthy—Principal, PSG College of Technology, Coimbatore

ICETEST_FM.indd 21 7/9/2018 4:25:53 PM

Author index 1035